Comparison of deep and conventional machine learning models for prediction of one supply chain management distribution cost | Scientific Reports

Scientific Reports volume 14, Article number: 24195 (2024) Cite this article

Metrics details

Strategic supply chain management (SCM) is essential for organizations striving to optimize performance and attain their goals. Prediction of supply chain management distribution cost (SCMDC) is one branch of SCM and it’s essential for organizations striving to optimize performance and attain their goals. For this purpose, four machine learning algorithms, including random forest (RF), support vector machine (SVM), multilayer perceptron (MLP) and decision tree (DT), along with deep learning using convolutional neural network (CNN), was used to predict and analyze SCMDC. A comprehensive dataset consisting of 180,519 open-source data points was used for analyze and make the structure of each algorithm. Evaluation based on Root Mean Square Error (RMSE) and Correlation coefficient (R2) show the CNN model has high accuracy in SCMDC prediction than other models. The CNN algorithm demonstrated exceptional accuracy on the test dataset, with an RMSE of RMSE of 0.528 and an R2 value of 0.953. Notable advantages of CNNs include automatic learning of hierarchical features, proficiency in capturing spatial and temporal patterns, computational efficiency, robustness to data variations, minimal preprocessing requirements, end-to-end training capability, scalability, and widespread adoption supported by extensive research. These attributes position the CNN algorithm as the preferred choice for precise and reliable SCMDC predictions, especially in scenarios requiring rapid responses and limited computational resources.

Supply Chain Management (SCM) refers to the strategic coordination and management of all activities involved in the sourcing, procurement, production, and distribution of goods and services1,2,3,4. It encompasses the entire process from the acquisition of raw materials to the delivery of the final product to the consumer5,6,7,8. The primary objective of SCM is to optimize the flow of products, information, and finances across the supply chain to achieve efficiency, reduce costs, and improve customer satisfaction9,10,11,12. By integrating and managing the supply chain as a cohesive system, organizations can enhance their competitiveness and responsiveness to market demands, ultimately leading to increased profitability and business performance12,13,14,15,16.

Despite the advancements in SCM, a significant gap exists in effectively managing and optimizing distribution costs17,18,19. Supply chain management distribution cost (SCMDC) is a critical component of the total supply chain cost, encompassing expenses related to transportation, warehousing, handling, and delivery of products20,21. The gap arises from the complexity and variability inherent in supply chain networks, which are influenced by factors such as fluctuating demand, transportation inefficiencies, and dynamic market conditions22. Traditional methods of cost estimation often fail to capture these nuances, leading to suboptimal decision-making and increased costs. Closing this gap requires a deeper understanding of the various cost drivers and the development of more sophisticated models that can accurately predict and manage distribution costs in real-time.

In recent years, the integration of advanced machine and deep learning models into SCM and other fields has become increasingly pivotal in addressing the complexities and inefficiencies traditionally associated with distribution cost prediction23,24,25,26. The sheer volume and diversity of data generated within supply chains, from procurement to delivery, present a significant challenge to conventional analytical approaches, which often fall short in identifying subtle patterns and relationships27,28. Machine learning, particularly deep learning, has emerged as a powerful tool capable of processing and analyzing this data to uncover intricate, non-linear relationships between various factors influencing distribution costs25,29.

This capability not only enhances the accuracy of predictions but also enables more informed and timely decision-making, ultimately leading to better cost management and resource allocation30,31. Moreover, the real-time predictive power of machine learning allows organizations to swiftly respond to fluctuations in demand, transportation costs, and other dynamic variables, reducing the risk of bottlenecks and inefficiencies32,33. The historical evolution of SCM has seen the development of numerous theories and methodologies aimed at achieving optimal efficiency, yet it is the advent of artificial intelligence that has truly revolutionized the field34. Researchers today are increasingly focused on refining these machine learning models, exploring their application across various SCM domains, and pushing the boundaries of what is possible in predictive analytics.

El-Khchine et al. (2018) explored the integration of machine learning techniques like K-Nearest Neighbors (KNN), Logistic Regression (LR), and Support Vector Machines (SVM) with Twitter data to enhance chicken SCM. Their approach identified consumer concerns, facilitating a consumer-centric supply chain design35. Bousqaoui et al. (2019) investigated the application of machine learning algorithms in supply chain processes, emphasizing their ability to improve predicting accuracy. They utilized a Long Short-Term Memory (LSTM) model to predict daily demand in a Moroccan supermarket, showcasing the potential of ML in SCM36. Islam and Amin (2020) used tree-based machine learning, specifically Distributed Random Forest (RF) and Gradient Boosting Machine (GBM), to predict product backorders in business decision processes. Their ranged approach improved model performance by 20%, offering flexibility and clarity while handling real-time data errors37. Alnahhal et al. (2021) utilized linear and logistic regression for dynamic lead-time predicting in make-to-order supply chains. Results showed reasonable accuracy, with an average type I error of 0.07, pioneering optimization in shipment temporal consolidation38. Oyewola et al. (2022) aimed to address challenges in supply chain management, such as lead times, bottlenecks, and quality assurance, by classifying supply chain pricing datasets of health medications. They employed deep learning techniques, specifically Long Short-Term Memory (LSTM) and One Dimensional Convolutional Neural Network (1D-CNN), alongside Bayesian optimization and All K Nearest Neighbor (AllkNN). The results demonstrated that the combination of 1D-CNN, AllkNN, and Bayesian optimization outperformed other models, achieving an accuracy range of 61.2836–63.3267%39. Al Moteri et al. (2023) aimed to enhance supply chain logistics operations by developing a novel strategy for estimating a macroeconomic index. The method used included multiple regression analysis (MRA) and adaptive extreme learning machine (AELM) models, combined with enhanced genetic algorithms and mathematical modeling. The results showed excellent and stable prediction accuracy, indicating the method’s potential usefulness. The conclusion highlighted the approach’s promise in improving cost-efficiency and economic value for businesses40. Taghiyeh et al. (2023) introduced a novel approach for hierarchical time series predicting in supply chains, leveraging machine learning techniques. By predicting child-level demands independently and then aggregating them, they achieved an 82–90% improvement in predict accuracy compared to traditional methods. This approach promises significant cost reductions in logistics, particularly beneficial for e-commerce operations41. Kim et al. (2024) aimed to identify effective machine learning technologies for managing the biodiesel supply chain to reduce operational costs. The study utilized a review of the scientific literature, focusing on various machine learning algorithms. The results highlighted that RF and Artificial Neural Networks (ANN) were the most accurate for predicting feedstock yield, biodiesel productivity, and quality. The conclusion emphasized their utility for engineers and managers in optimizing supply chain operations42. Alshurideh et al. (2024) aimed to enhance the transparency and integrity of supply chains by addressing the vulnerabilities of centralized Supply Chain Management (SCM) systems. The study employed a blockchain-based supply chain management model, integrating Machine Learning (ML) techniques. The results demonstrated significant improvements in product distribution, traceability, partner cooperation, and financing access. The authors concluded that blockchain, combined with ML, can substantially improve SCM performance in the business sector43. Amellal et al. (2024) aimed to enhance strategic decision-making in businesses by improving the interpretation of customer sentiment, demand forecasting, and price prediction. The study employed a comprehensive methodology, including the BERT transformer model for sentiment analysis, the Gated Recurrent Unit (GRU) model for demand forecasting, and the Bayesian Network for price prediction. The results indicated superior performance over traditional methods, concluding that this integrative approach provides valuable insights for optimizing pricing strategies and managing supply chain uncertainties44.

This article discusses a gap in the supply chain field related to shipping and its impact on predicting fluctuations in SCM distribution cost (SCMDC) across various transportation methods by different variables. Its objective is to address this issue by using algorithms, particularly artificial intelligence (AI) and deep learning. The study utilized data from 180,519 open-source datasets and used algorithms such as Random Forest (RF), Support Vector Machine (SVM), Multi-Layer Perceptron (MLP), Decision Tree (DT), and Convolutional Neural Network (CNN) to predict SCMDC of goods (clothing, sports, and electronic supplies), including freight costs and other features (payment type, scheduled shipment day, actual shipment day, late delivery risk, order item discount, order item discount rate, order item profit ratio, order item quantity, order profit per order, shipping mode, sales per customer, and order item product price), for transportation to their destination. Finally, the study compares the efficacy of AI and deep learning algorithms, proposing a method for managing supply chain pricing accordingly.

This article uses the conventional algorithms such as RF, SVM, MLP, DT, and the deep learning algorithm such as CNN to predict the SCMDC within the supply chain.

Random Forest (RF) algorithm, a widely utilized and potent technique in machine learning, is used for diverse tasks like classification and prediction45. This approach entails amalgamating multiple randomly selected decision trees, whose outcomes are then aggregated to yield the final result46. The algorithm operates by first selecting random data samples for each decision tree, which may or may not involve replacement. Subsequently, decision trees are constructed based on these samples and their associated labels. Upon completion, new data samples are introduced to each tree for prediction, and the resulting predictions are combined, often through majority voting or averaging47. The model’s performance is then assessed using metrics like accuracy and sensitivity. Noteworthy advantages include its robustness against overfitting, enhanced prediction accuracy through result combination, and capability to handle large, high-dimensional datasets. However, drawbacks include increased time and memory requirements compared to standard decision trees, as well as potentially time-consuming parameter configuration48. Further details on RF are available in Barjouei et al. (2021)49. Figure 1, show the graphical diagram for RF.

Block diagram for RF algorithm.

Support vector machine (SVM), introduced by Cortes and Vapnik 1995, stands as one of the most impactful machines learning techniques, effectively addressing numerous regression challenges50. Its core principle relies on statistical learning and structural risk minimization to diminish empirical risk and facilitate generalization51. The SVM operates by transforming input data nonlinearly into a higher-dimensional feature space, followed by the application of a kernel to tackle linear regression within this transformed space52. Further details on SVM are available in Abad et al. (2021)53. This study specifically utilizes Polynomial kernels within the SVM framework. Figure 2, show the graphical diagram for SVM.

Block diagram for SVM algorithm.

The ANN have been extensively utilized since the 1990s54. Several factors affect their predictive accuracy, including feature selection, network architecture, transfer functions, and training algorithm choice55. The Multilayer Perceptron (MLP) is commonly used due to its adaptability56. However, the Levenberg-Marquardt (LM) algorithm used for MLP training may face limitations with complex datasets57. Further details on MLP are available in Taud and Mas (2018)58. Using more effective optimization algorithms is important for enhancing MLP performance. Sensitivity analysis often recommends a two-hidden-layer structure, with 10 neurons in hidden layer 1 and 5 in hidden layer 2. Transfer functions ‘tansig’ and ‘purelin’ are typically chosen for these layers. Figure 3, show the graphical diagram for MLP.

Block diagram for MLP algorithm.

Decision trees (DT) are renowned as a potent technique used across various domains such as machine learning, image processing, and pattern recognition. This algorithm operates as a hierarchical model, making sequential decisions based on multiple tests conducted on input features59. A notable aspect of decision trees is their straightforward structure, facilitating the interpretation of rules and criteria, contrasting with methods like neural networks relying on numerical weights60. The primary application of decision trees lies in data mining and information classification and prediction61. In this model, each node of the tree embodies certain characteristics directly influencing the decision-making process for data classification. By establishing a hierarchical arrangement of nodes and branches, decision trees proficiently analyze complex data. Given their simplicity and accuracy in data analysis, decision trees have garnered attention and find application across diverse domains, including data analysis and classification problem-solving. These algorithms serve as robust and adaptable tools in data analysis. Further details on DT are available in Kamali et al. (2022)62. The Fig. 4 diagram below illustrates an example of a decision tree’s structure.

Block diagram for DT algorithm.

Convolutional neural networks (CNNs) offer distinct advantages in extracting image features, with parameter sharing serving as the cornerstone for processing input images of diverse sizes63. A CNN typically comprises convolution layers, pooling layers, and fully connected layers, with the convolution layer serving as its core64. In this layer, the same weight matrix and bias matrix are used to compute inputs across varying positions (Eq. 1):

where the Mi, j is the output factor; bi, n, Si, n and Wi, n are bais, matrix in specific position and weight of the matrix.

The convolutional process entails sequentially applying the convolution kernel to different locations within the image field, often accompanied by edge-padding operations to ensure dimensional consistency between input and output fields65. Compared to conventional fully connected neural networks, utilizing a convolution kernel with shared parameters can streamline model complexity and enhance efficiency66,67. Furthermore, using uniform parameters throughout the image field overcomes localized specificity limitations, thereby unveiling hidden rules applicable to each location. Through training, the model dynamically adjusts the convolution kernel’s parameters to learn spatial connections, markedly enhancing the model’s spatial generalization capability68,69. Nonetheless, the convolution process may not be universally applicable to all types of model inputs. Data such as air temperature, cloud cover, and wind speed exhibit relatively stable spatial distributions within a given area, often devoid of spatial distribution monitoring data, and are hence uniformly distributed in space with only temporal variations considered. However, this algorithm demonstrates superior accuracy when handling complex input variables with disparate data distributions70. This algorithm boasts numerous advantages, including automated feature extraction, facilitated by CNNs’ ability to automatically discern essential features from data without human intervention. Parameter sharing reduces the computational burden by using shared parameters across the network71. With the help of pooling layers, CNNs can detect meaningful patterns in variable data based on local features. Additionally, CNNs exhibit robust performance in recognizing intricate patterns and processing large-dimensional images owing to their inherent structural adaptability. Moreover, CNNs showcase flexibility in accommodating various inputs, encompassing images of diverse sizes and dimensions, thereby rendering them applicable to a myriad of problems. Figure 5 for a visual representation of a CNN architecture. The CNN architecture designed with 12 input variables and one output variable includes the following detailed technical specifications: The network begins with an input layer, followed by two convolutional layers with 64, and 128 filters, respectively, each with a 3 × 3 kernel size and ReLU activation functions. Max-pooling layers with a 2 × 2 pool size follow each convolutional layer to reduce dimensionality. Batch normalization is applied after each convolution to stabilize learning. The architecture includes dropout layers with a rate of 0.5 after the second convolutional layer and before the fully connected layer to prevent overfitting. A fully connected layer with 256 neurons maps the features to the output, with a softmax activation function applied in the output layer to generate the final prediction. Key hyperparameters include a learning rate of 0.001, a batch size of 32, and a total of 50 epochs for training.

Block diagram for CNN algorithm.

One of the methodologies utilized for data validation involves using the k-fold cross-validation technique, widely recognized as among the most effective approaches72. This technique involves treating the dataset as a cohesive entity and segmenting it into multiple subsets. Initially, a portion of the data is earmarked as test data, while the remainder serves as training data73. Subsequently, these roles are reversed, with a different subset of data designated as the test set. This iterative process is reiterated for each of the k partitions, typically set at 7 in this context. Generally, the procedure is repeated 10 times, resulting in an average of 70 iterations (Fig. 6).

By using this validation technique, various data analysis challenges are tackled, bolstering the reliability of outcomes while mitigating issues like overfitting and algorithmic inefficiencies in prediction74. In this methodology, one subset of data functions as the test set, while the remaining 6 subsets serve as training data. Ultimately, the average of the minimum values derived from the data serves as the measure of prediction accuracy. This iterative approach substantially contributes to result validation, ensuring the robustness of outcomes.

Block diagram K-fold cross validation.

The study used data from 180,519 open-source datasets (https://data.mendeley.com/datasets/8gx2fvg2k6/5) and utilized algorithms such as RF, SVM, MLP, DT, and CNN to predict the distribution cost of goods (clothing, sports, and electronic supplies), incorporating factors such as payment type, scheduled shipment day, actual shipment day, late delivery risk, order item discount, order item discount rate, order item profit ratio, order item quantity, order profit per order, shipping mode, sales per customer, and order item product price. Payment types included debit, transfer, and cash, while shipping modes consisted of standard class, first class, second class, and same-day delivery. Table 1 presents statistical analysis for the input variables and predicted SCMDC.

In this article, the prediction of the supply chain concerning SCMDC transportation is explored through the application of deep and machine learning algorithms. A novel approach is undertaken by using a model from the Open Data dataset, representing pioneering work in this area where limited research exists. Thus, this article presents a novelty as artificial intelligence methods have not been previously utilized. Comparing various algorithms for predicting this crucial aspect involves the use of statistical parameters for comparison. To evaluate the machine learning algorithms shown in the Eqs. 2–6.

Table 2 presents a comparison of different machine learning (ML) and deep learning (DL) algorithms for SCMDC prediction. This article employs RF, SVM, MLP, DT, and CNN algorithms for this purpose, utilizing 180,519 open-source datasets. 70% of the dataset is allocated for training, 15% for testing, and the remaining 15% for validation. The results obtained guide an analysis of which approach is optimal for predicting product arrival, aiding marketing experts in optimizing product delivery to customers at minimal cost disruptions, thereby enhancing producers’ profits.

Based on the data presented in Tables 2, 3 and 4, it is evident that the CNN algorithm outperforms other algorithms in terms of accuracy. Specifically, for the test data detailed in Tables 2, 3 and 4, the MLP algorithm exhibits MRE = 0.004, MARE = 6.328, STD = 8.589, RMSE = 2.715, and R2 = 0.880. Similarly, the DT algorithm shows MRE = 0.009, MARE = 4.004, STD = 6.288, RMSE = 1.871, and R2 = 0.888, the RF algorithm shows MRE = 0.003, MARE = 1.329, STD = 1.675, RMSE = 0.528, and R2 = 0.953 and the SVM algorithm shows MRE=-0.068, MARE = 6.202, STD = 7.836, RMSE = 2.546, and R2 = 0.886. The CNN algorithm also displays MRE=-0.030, MARE = 5.882, STD = 8.166, RMSE = 2.492, and R2 = 0.883, indicating its superior accuracy compared to the RF, SVM, MLP, DT algorithm, which also presents similar metrics. Consequently, the CNN algorithm proves to have higher accuracy than other artificial intelligence algorithms utilized in this study.

Figure 7 displays a cross-plot illustrating the relationship between predicted and measured data. This visual representation reveals the superior performance of the CNN deep learning algorithm compared to RF, SVM, MLP, and DT machine learning algorithms. Notably, the CNN algorithm achieves an R2 value of 0.953, indicating a strong correlation between predicted and measured data points. Comparison of R2 values across algorithms in this study establishes their performance accuracy hierarchy as follows: CNN > DT > RF > SVM > MLP. These findings corroborate earlier observations from Table 2, where the CNN algorithm demonstrated superior predictive capabilities for SCMDC compared to RF, SVM, MLP, and DT algorithms.

Cross plot predicting SCMDC using three DL and ML algorithms.

Figure 8 visually illustrates the computational error for test data, depicting the error distribution in SWE prediction across RF, SVM, MLP, DT, and CNN algorithms. The plotted coordinates in the figure delineate the error range for each algorithm, with DT, MLP, SVM, RF, and CNN exhibiting error ranges from − 10.4 to 10.5, -17.9 to 17.9, -14.3 to 11.9, -33.2 to 11.5, and − 6.8 to 10.3, respectively. This data illustrates that the CNN model’s predictions for test data exhibit a comparatively minor deviation from actual SCMDC values within this range. Consequently, the DT algorithm shows smaller errors, followed by the RF algorithm, while the SVM and MLP algorithms demonstrate larger error ranges. These results reinforce the conclusion that the CNN algorithm surpasses the RF, SVM, MLP, and DT algorithms in SCMDC prediction accuracy, consistently generating predictions with lesser errors and narrower error margins.

Error data point based predicting SCMDC using three DL and ML algorithms.

Figure 9 illustrates the distribution of prediction errors for the RF, SVM, MLP, DT, and CNN algorithms in SCMDC prediction. Each graph displays error distribution, ideally normal with a mean at zero and relatively low dispersion without significant deviations. These graphs are useful for analyzing algorithm performance and identifying the best one based on error distribution. Closer examination reveals that the CNN algorithm offers superior accuracy compared to others. It is characterized by a smaller standard deviation and narrower dispersion of prediction errors, indicating its more effective data prediction capability. This result demonstrates that the CNN algorithm consistently provides more accurate and reliable predictions for SCMDC. Based on the results from Tables 2, 3 and 4 and the analysis of Fig. 9, the algorithms rank in performance accuracy as follows: CNN > DT > RF > SVM > MLP.

Error histogram based predicting SCMDC using three DL and ML algorithms.

The Correlation coefficient (R) assesses the significance of dependent and independent variables in systems like SCMDC models. R ranges from − 1 to + 1, indicating correlation strength and direction. Values near 1 signify strong positive correlation, near − 1 indicate strong negative correlation, and close to zero imply no correlation. Equation 7 computes the efficiency coefficient, quantifying the linear relationship between variables. It aids researchers in gauging the impact of independent variables on output within the SCMDC model.

A + 1 correlation indicates a perfect positive correlation, implying the most positive effect of independent variables on dependent ones. Conversely, a -1 correlation suggests a complete negative influence of independent variables. Near-zero correlation indicates no significant relationship, implying minimal impact of independent variables on dependents. The coefficient of performance quantifies relationships, assessing the relative importance of independent variables in the SCMDC model. Using the heatmap shown in Fig. 10, a comparison of Pearson’s correlation coefficients provides insight into the relationships between the input variables and SCMDC. The results reveal several significant correlations among the variables. Negative correlations are observed with payment type, late delivery risk, order item profit ratio, and shipping mode, indicating an inverse relationship with SCMDC. Conversely, positive correlations are found with the shipment’s day (scheduled), shipment’s day (real), order item discount, order item discount rate, order item quantity, order profit per order, sales per customer, and order item product price. These variables demonstrate a direct relationship with SCMDC. Notably, the high values of parameters such as order item discount, order item discount rate, sales per customer, and order item product price suggest that these factors have a substantial impact on SCMDC.

Heat map block diagram to input/output variables prediction of SCMDC.

While the study effectively demonstrates the superiority of CNN over conventional machine learning models for predicting SCMDC, it also presents some limitations. The research relies heavily on a single dataset, which may not fully capture the variability and complexity of different supply chains. This could limit the generalizability of the findings to other contexts. Although traditional machine learning algorithms such as SVM, RF, MLP, and DT offer robust capabilities for prediction tasks, they present certain limitations when applied to complex datasets like those in SCMDC prediction. These algorithms often require extensive manual feature engineering to perform effectively, which can be time-consuming and may overlook intricate patterns within the data. Additionally, they may struggle with capturing non-linear relationships and interactions between features, leading to less accurate predictions compared to deep learning models like CNNs. Furthermore, conventional algorithms are sometimes less adaptable to the large-scale, high-dimensional datasets typical in SCM, and their performance may degrade as the complexity of the data increases.

Only proofread: Today, Artificial Intelligence (AI) has advanced to the point where it can replace traditional methods and techniques, addressing human challenges in increasingly sophisticated ways. Its progress suggests that AI will become a prominent tool in various fields in the near future. This includes determining and exploring other key techniques and methods that have been discussed in diverse domains, particularly those that could benefit from approaches such as CNN, SVM, RF, MLP, and DT. For example, for future work in the field of supply chain management for future, these AI methods have significant potential for application and comparison like these works50,52,75,76,77. Similarly, they can be beneficial in energy where they might optimize processes and improve efficiency like these works48,78,79,80,81. Moreover, in electrical engineering, these techniques could be integrated to address complex challenges and enhance system performance like these works82,83,84,85,86. In mechanical engineering, a fundamental approach using these methods could offer solutions to problems currently faced by professionals like these works87,88,89,90. However, it is important to note that solving such complex issues may require substantial time.

Supply chain management (SCM) is crucial for achieving organizational success and advancing towards strategic goals. Effective decision-making in financial chain management is essential for analyzing acquired data, reducing supply chain management distribution costs (SCMDC), and maximizing profits. This study aims to improve efficiency and accuracy in SCM through SCMDC forecasting by employing four machine learning algorithms—Random Forest (RF), Support Vector Machine (SVM), Multi-Layer Perceptron (MLP), and Decision Tree (DT)—alongside deep learning with Convolutional Neural Network (CNN). A comprehensive dataset consisting of 180,519 open-source data points was divided into 70% for training, 15% for testing, and 15% for validation. The evaluation of SCMDC prediction performance, based on Root Mean Square Error (RMSE), ranked the models as follows: CNN > DT > RF > SVM > MLP. The CNN model demonstrated exceptional accuracy in SCMDC prediction on the test dataset, achieving an RMSE of 0.528 and an R² value of 0.953. CNNs are particularly noted for their robustness and efficiency in handling large datasets, as they can automatically learn hierarchical features, capture spatial and temporal patterns effectively, and maintain computational efficiency through weight sharing. Their ability to handle intricate datasets, minimal preprocessing requirements, and end-to-end training capability make CNNs a superior choice for accurate and reliable SCMDC predictions. Looking forward, future research could explore integrating CNNs with emerging technologies such as real-time data analytics and advanced optimization techniques to further enhance SCM efficiency and adapt to evolving challenges in supply chain dynamics.

To further enhance SCM efficiency, future research should focus on integrating Convolutional Neural Networks (CNNs) with real-time data analytics and advanced optimization techniques. Exploring hybrid models that combine CNNs with reinforcement learning or other AI methodologies could improve adaptability to dynamic supply chain environments. Additionally, incorporating diverse datasets and expanding research to different industry contexts may offer deeper insights and more robust SCMDC prediction capabilities.

Data accessible upon academic request from corresponding authors.

TiAdaptive extreme learning machineme

Artificial neural networks

Artificial intelligence

Convolutional neural network

Deep learning

Distributed random forest

Decision tree

Gradient boosting machine

Gated recurrent unit

K-nearest neighbors

Logistic regression

Long short-term memory

Machine learning

Multi-layer perceptron

Multiple regression analysis

Correlation coefficient

Random forest

Supply chain management

Supply chain management distribution cost

Support vector machines

Omigie, S. & Oguns-Obasohan, I. The the influence of supply chain management practises on operational performance of quoted manufacturing firms in Nigeria: Procurement outsourcing and order process management view. J. Entrepreneurship Bus. 11(1), 89–104 (2023).

Article Google Scholar

Enz, M. G. & Lambert, D. M. A supply chain management framework for services. J. Bus. Logist. 44(1), 11–36 (2023).

Article Google Scholar

Fan, W., Wu, X. & He, Q. Digitalization drives green transformation of supply chains: A two-stage evolutionary game analysis 1–20 (Annals of Operations Research, 2024).

Google Scholar

He, B. & Yin, L. Prediction modelling of cold chain logistics demand based on data mining algorithm. Math. Probl. Eng. 2021(1), 3421478 (2021).

Google Scholar

Xiao, X. et al. Improving traceability and transparency of table grapes cold chain logistics by integrating WSN and correlation analysis. Food Control. 73, 1556–1563 (2017).

Article Google Scholar

Taleizadeh, A. A. et al. New products or remanufactured products: Which is consumer-friendly under a closed-loop multi-level supply chain?. J. Retailing Consumer Serv. 73, 103295 (2023).

Article Google Scholar

Xu, X. & Wei, Z. Dynamic pickup and delivery problem with transshipments and LIFO constraints. Comput. Ind. Eng. 175, 108835 (2023).

Article Google Scholar

Luo, J. et al. Transport infrastructure connectivity and conflict resolution: A machine learning analysis. Neural Comput. Appl. 34(9), 6585–6601 (2022).

Article Google Scholar

Lee, R. The effect of supply chain management strategy on operational and financial performance. Sustainability. 13(9), 5138 (2021).

Article Google Scholar

Rahiminezhad Galankashi, M., Mokhatab, F. & Rafiei,. Financial performance measurement of supply chains: A review. Int. J. Product. Perform. Manage. 71(5), 1674–1707 (2022).

Article Google Scholar

Ma, K. et al. Demand-side energy management considering price oscillations for residential building heating and ventilation systems. IEEE Trans. Ind. Inf. 15(8), 4742–4752 (2019).

Article Google Scholar

Ning, F. et al. Manufacturing cost estimation based on similarity. Int. J. Comput. Integr. Manuf. 36(8), 1238–1253 (2023).

Article Google Scholar

Madhani, P. M. Customer-focused supply chain strategy: Developing 4Rs framework for enhancing competitive advantages. Int. J. Serv. Oper. Manage. 36(4), 505–530 (2020).

Google Scholar

Richey, R. G. et al. A responsiveness view of logistics and supply chain management. J. Bus. Logist. 43(1), 62–91 (2022).

Article MathSciNet Google Scholar

Mukhtar, U. & Azhar, T. Inter-functional coordination to co-create value within integrated value chains for competitive supply chain. Int. J. Oper. Supply Chain Manag. 13(1), 11–22 (2020).

Google Scholar

Ma, K., Yang, J. & Liu, P. Relaying-assisted communications for demand response in smart grid: Cost modeling, game strategies, and algorithms. IEEE J. Sel. Areas Commun. 38(1), 48–60 (2019).

Article CAS Google Scholar

Wu, L. et al. Smart supply chain management: A review and implications for future research. Int. J. Logist. Manage. 27(2), 395–417 (2016).

Article Google Scholar

Soni, G. et al. Swarm intelligence approaches in supply chain management: Potentials, challenges and future research directions. Int. J. Supply Chain Manag. 24(1), 107–123 (2019).

Article Google Scholar

Rehan, M. et al. Supply Chain Management Using an Industrial Internet of Things Hyperledger Fabric Network13 (Human-centric computing and information sciences, 2023).

Whitehead, T. A. Investigatory Study into the Supply Chain Voids of Welsh Industrial Sectors (Cardiff University United Kingdom, 2010).

Lu, K. Online distributed algorithms for online noncooperative games With stochastic cost functions: High probability bound of regrets. IEEE Trans. Autom. Control (2024).

Jain, V. & Benyoucef, L. Managing long supply chain networks: Some emerging issues and challenges. J. Manuf. Technol. Manage. 19(4), 469–496 (2008).

Article Google Scholar

Xu, A., Li, Y. & Donta, P. K. Marketing decision model and consumer behavior prediction with deep learning. J. Organ. End. User Comput. (JOEUC) 36(1), 1–25 (2024).

Article CAS Google Scholar

Yin, L. et al. AFBNet: A Lightweight Adaptive Feature Fusion Module for Super-Resolution Algorithms 140 (CMES-Computer Modeling in Engineering & Sciences, 2024). 3.

Tirkolaee, E. B. et al. Application of machine learning in supply chain management: A comprehensive overview of the main areas. Math. Probl. Eng. 2021(1), 1476043 (2021).

Google Scholar

Fu, X. et al. Tolerance analysis of cyber-manufacturing systems to cascading failures. ACM Trans. Internet Technol. 23(4), 1–23 (2023).

Article CAS Google Scholar

Gattorna, J. Dynamic Supply Chains: How to Design, Build and Manage People-Centric Value Networks (FT, 2015).

Google Scholar

Carbonneau, R., Laframboise, K. & Vahidov, R. Application of machine learning techniques for supply chain demand forecasting. Eur. J. Oper. Res. 184(3), 1140–1154 (2008).

Article Google Scholar

Xin, J. et al. A deep-learning-based MAC for integrating channel access, rate adaptation and channel switch. arXiv preprint arXiv:2406.02291, 2024.

Koot, M., Mes, M. R. K. & Iacob, M. E. A systematic literature review of supply chain decision making supported by the internet of things and Big Data Analytics. Comput. Ind. Eng. 154, 107076 (2021).

Article Google Scholar

Priore, P. et al. Applying machine learning to the dynamic selection of replenishment policies in fast-changing supply chain environments. Int. J. Prod. Res. 57(11), 3663–3677 (2019).

Article Google Scholar

Aljohani, A. Predictive analytics and machine learning for real-time supply chain risk mitigation and agility. Sustainability. 15(20), 15088 (2023).

Article Google Scholar

Younis, H., Sundarakani, B. & Alsharairi, M. Applications of artificial intelligence and machine learning within supply chains: Systematic review and future research directions. J. Modell. Manage. 17(3), 916–940 (2022).

Article Google Scholar

Liu, K. S. & Lin, M. H. Performance assessment on the application of artificial intelligence to sustainable supply chain management in the construction material industry. Sustainability. 13(22), 12767 (2021).

Article Google Scholar

El-Khchine, R. et al. Machine Learning for Supply Chain’s Big Data: State of the art and Application to Social Networks’ data (EDP Sciences, 2018).

Bousqaoui, H., Achchab, S. & Tikito, K. Machine Learning Applications in Supply Chains: Long short-term Memory for Demand Forecasting (Springer, 2019).

Islam, S. & Amin, S. H. Prediction of probable backorder scenarios in the supply chain using Distributed Random Forest and Gradient Boosting Machine learning techniques. J. Big Data. 7(1), 65 (2020).

Article Google Scholar

Alnahhal, M., Ahrens, D. & Salah, B. Dynamic lead-time forecasting using machine learning in a make-to-order supply chain. Appl. Sci. 11(21), 10105 (2021).

Article CAS Google Scholar

Oyewola, D. O. et al. Application of deep learning techniques and bayesian optimization with tree parzen estimator in the classification of supply chain pricing datasets of health medications. Appl. Sci. 12(19), 10166 (2022).

Article CAS Google Scholar

Al Moteri, M., Khan, S. B. & Alojail, M. Economic growth forecast model urban supply chain logistics distribution path decision using an improved genetic algorithm. Malaysian J. Comput. Sci., pp. 76–89. (2023).

Taghiyeh, S. et al. A novel multi-phase hierarchical forecasting approach with machine learning in supply chain management. Supply Chain Analytics. 3, 100032 (2023).

Article Google Scholar

Kim, S., Seo, J. & Kim, S. Mach. Learn. Technol. Supply Chain Manage. Res. Biodiesel: Rev. Energies, 17(6): 1316 (2024).

CAS Google Scholar

Alshurideh, M. T. et al. Empowering Supply Chain Management System with Machine Learning and Blockchain Technology, in Cyber Security Impact on Digitalization and Business Intelligence: Big Cyber Security for Information Management: Opportunities and Challenges p. 335–349 (Springer, 2024).

Amellal, I. et al. An integrated approach for modern supply chain management: Utilizing advanced machine learning models for sentiment analysis, demand forecasting, and probabilistic price prediction. Decis. Sci. Lett. 13(1), 237–248 (2024).

Article Google Scholar

Denisko, D. & Hoffman, M. M. Classification and interaction in random forests. Proc. Nati .Acad. Sci., 115(8), 1690–1692 (2018).

Yin, L. et al. PriMonitor: an adaptive tuning privacy-preserving approach for multimodal emotion detection. World Wide Web. 27(2), 9 (2024).

Article Google Scholar

Yin, L. et al. A game-theoretic approach for federated learning: a trade-off among privacy, accuracy and energy. Digit. Commun. Networks. 10(2), 389–403 (2024).

Article MathSciNet Google Scholar

Shi, S., Han, D. & Cui, M. A multimodal hybrid parallel network intrusion detection model. Connection Sci. 35(1), 2227780 (2023).

Article Google Scholar

Barjouei, H. S. et al. Prediction performance advantages of deep machine learning algorithms for two-phase flow rates through wellhead chokes. J. Petroleum Explor. Prod. 11(3), 1233–1261 (2021).

Article CAS Google Scholar

Zhou, J. et al. Impacts of IT capability and supply chain collaboration on supply chain resilience: Empirical evidence from China in COVID-19 pandemic. J. Enterp. Inform. Manage. 37(2), 777–803 (2024).

Article Google Scholar

Wang, Z. et al. Investigating failure patterns in machine learning-based object detection tasks in Software Development Courses. J. Internet Technol. 24(4), 1001–1008 (2023).

Article Google Scholar

Guo, X. et al. Research on network security situation awareness and dynamic game based on deep Q learning network. J. Internet Technol. 24(2), 549–563 (2023).

Google Scholar

Abad, A. R. B. et al. Predicting oil flow rate through orifice plate with robust machine learning algorithms. Flow Meas. Instrum. 81, 102047 (2021).

Article Google Scholar

Misra, J. & Saha, I. Artificial neural networks in hardware: A survey of two decades of progress. Neurocomputing. 74(1–3), 239–255 (2010).

Article Google Scholar

Maier, H. R. & Dandy, G. C. Neural Networks for the Prediction and Forecasting of Water Resources Variables: A Review of Modelling Issues and Applications 15p. 101–124 (Environmental modelling & software, 2000). 1.

Abiodun, O. I. et al. Comprehensive review of artificial neural network applications to pattern recognition. IEEE Access. 7, 158820–158846 (2019).

Article Google Scholar

Heidari, A. A. et al. Ant lion optimizer: Theory, literature review, and application in multi-layer perceptron neural networks. Nature-Inspired Optimizers: Theories, Literature Reviews and Applications, pp. 23–46 (2020).

Taud, H. & Mas, J. F. Multilayer perceptron (MLP). Geomatic approaches for modeling land change scenarios, pp. 451–455 (2018).

Xia, Y. et al. A boosted decision tree approach using bayesian hyper-parameter optimization for credit scoring. Expert Syst. Appl. 78, 225–241 (2017).

Article Google Scholar

Browne, A. et al. Biological data mining with neural networks: Implementation and application of a flexible decision tree extraction algorithm to genomic problem domains. Neurocomputing. 57, 275–293 (2004).

Article Google Scholar

Maimon, O. Z. & Rokach, L. Data Mining with Decision Trees: Theory and Applications Vol. 81 (World scientific, 2014).

Kamali, M. Z. et al. Permeability prediction of heterogeneous carbonate gas condensate reservoirs applying group method of data handling. Mar. Pet. Geol. 139, 105597 (2022).

Article Google Scholar

Mantey, E. A. et al. Maintaining Privacy for a Recommender System Diagnosis Using Blockchain and deep Learning 13 (Human-centric computing and information sciences, 2023).

Malibari, A. A. et al. Metaheuristics with Deep Learning Enabled Epileptic Seizure Classification for Smart Healthcare on Cyborg Robots 13 (Human-centric Computing and Information Sciences, 2023).

Carranza-García, M., García-Gutiérrez, J. & Riquelme, J. C. A framework for evaluating land use and land cover classification using convolutional neural networks. Remote Sens. 11(3), 274 (2019).

Article ADS Google Scholar

Capra, M. et al. An updated survey of efficient hardware architectures for accelerating deep convolutional neural networks. Future Internet. 12(7), 113 (2020).

Article Google Scholar

Ryu, S., Kim, J. & Park, N. Study on Trends and predictions of convergence in Cybersecurity Technology using machine learning. J. Internet Technol. 24(3), 709–725 (2023).

Article Google Scholar

Zhang, W. et al. Short-term traffic flow prediction based on spatio-temporal analysis and CNN deep learning. Transportmetrica A Transp. Sci.15(2), 1688–1711 (2019).

Article Google Scholar

Zhu, L. et al. A Novel Hybrid Excitation Magnetic lead Screw and its Transient sub-domain Analytical Model for wave Energy Conversion (IEEE Transactions on Energy Conversion, 2024).

Zhao, B. et al. Intelligent fault diagnosis of rolling bearings based on normalized CNN considering data imbalance and variable working conditions. Knowl. Based Syst. 199, 105971 (2020).

Article Google Scholar

Ang, K. M. et al. MTLBORKS-CNN: An innovative Approach for automated convolutional neural Network design for image classification. Mathematics. 11(19), 4115 (2023).

Article Google Scholar

Li, M. et al. Artificial intelligence enabled self-powered wireless sensing for smart industry. Chem. Eng. J. 492, 152417 (2024).

Article CAS Google Scholar

Wang, B. et al. Stacked noise reduction auto Encoder–OCEAN: A novel personalized recommendation model enhanced. Systems. 12(6), 188 (2024).

Article Google Scholar

Keevers, T. L. Cross-validation is Insufficient for Model Validation (Joint and Operations Analysis Division, 2019).

Liao, Z. et al. Distributed probabilistic offloading in edge computing for 6G-enabled massive internet of things. IEEE Internet Things J. 8(7), 5298–5308 (2020).

Article Google Scholar

Harju, A. et al. The impact of procurement digitalization on supply chain resilience: Empirical evidence from Finland. Supply Chain Manag. Int. J. 28(7), 62–76 (2023).

Article Google Scholar

Centobelli, P. et al. Sailing through a digital and resilient shipbuilding supply chain: An empirical investigation. J. Bus. Res. 158, 113686 (2023).

Article Google Scholar

Zhang, D. et al. A Multi-Source Dynamic Temporal Point Process Model for Train Delay Prediction (IEEE Transactions on Intelligent Transportation Systems, 2024).

Xinyu, W., Haoran, L. & Khan, K. Innovation in Technology: A game Changer for Renewable Energy in the European Union? (Wiley Online Library, 2024).

Li, J. et al. Concentrated solar power for a reliable expansion of energy systems with high renewable penetration considering seasonal balance. Renew. Energy. 226, 120089 (2024).

Article Google Scholar

Cheng, H. et al. Incentive-based demand response optimization method based on federated learning with a focus on user privacy protection. Appl. Energy. 358, 122570 (2024).

Article Google Scholar

Wang, J. et al. Intelligent ubiquitous network accessibility for wireless-powered MEC in UAV-assisted B5G. IEEE Trans. Netw. Sci. Eng. 8(4), 2801–2813 (2020).

Article Google Scholar

Cao, D. et al. BERT-based deep spatial-temporal network for taxi demand prediction. IEEE Trans. Intell. Transp. Syst. 23(7), 9442–9454 (2021).

Article Google Scholar

Liao, Z. et al. Blockchain on security and forensics management in edge computing for IoT: A comprehensive survey. IEEE Trans. Netw. Serv. Manage. 19(2), 1159–1175 (2021).

Article Google Scholar

Li, W. et al. Multimodel framework for indoor localization under mobile edge computing environment. IEEE Internet Things J. 6(3), 4844–4853 (2018).

Article Google Scholar

Li, W. et al. Complexity and algorithms for superposed data uploading problem in networks with smart devices. IEEE Internet Things J. 7(7), 5882–5891 (2019).

Article Google Scholar

Xie, X. et al. Fluid Inverse Volumetric Modeling and Applications from Surface Motion (IEEE Transactions on Visualization and Computer Graphics, 2024).

Tang, H. et al. Rational design of high-performance epoxy/expandable microsphere foam with outstanding mechanical, thermal, and dielectric properties. J. Appl. Polym. Sci. 141(24), e55502 (2024).

Article CAS Google Scholar

Wang, K. et al. Experimental study of mechanical properties of hot dry granite under thermal-mechanical couplings. Geothermics. 119, 102974 (2024).

Article Google Scholar

Dong, J. et al. Mechanical behavior and impact resistance of rubberized concrete enhanced by basalt fiber-epoxy resin composite. Constr. Build. Mater. 435, 136836 (2024).

Article CAS Google Scholar

Download references

This work was supported by the National Natural Science Foundation of China (No.62066032); Natural Science Foundation of Guangxi Province (No.2021GXNSFAA075019); The “14th Five Year Plan” of Guangxi Education and Science special project of college innovation and entrepreneurship education (No.2022ZJY2727); The “14th Five Year Plan” of Guangxi Education and Science Annual project in 2023 (No.2023A028); Middle-aged and Young Teachers’ Basic Ability of Scientific Research Promotion Project of Guangxi (No.2021KY0130); Philosophy and Social Science Foundation of Guangxi (No.21FYJ041). This study acknowledges the support of National First-class Undergraduate Major - The Major of Logistics Management, Demonstrative Modern Industrial School of Guangxi University - Smart Logistics Industry School Construction Project, the Logistics Engineering Innovation Laboratory, Logistics Engineering Technology Laboratory and Smart Logistics Exhibition Center of Nanning Normal University. The authors gratefully acknowledge the support of Construction project of Practice conditions and practice Base for industry-university cooperation of the Ministry of Education (No.202102079139).

Department of Logistics Management and Engineering, Nanning Normal University, Nanning, 530001, Guangxi, China

Xiaomo Yu

College of The Arts, Guangxi Minzu University, Nanning, 530001, Guangxi, China

Ling Tang

College of Computer Science and Information Engineering, Nanning Normal University, Nanning, 530001, Guangxi, China

Long Long

Department of Petroleum Engineering, Omidiyeh Branch, Islamic Azad University, Omidiyeh, Iran

Mohammad Sina

You can also search for this author in PubMed Google Scholar

You can also search for this author in PubMed Google Scholar

You can also search for this author in PubMed Google Scholar

You can also search for this author in PubMed Google Scholar

Conceptualization, XY, LT, LL and MS; formal analysis, LT and MS; investigation, XY, LT, LL and MS; methodology, XY and MS; project administration, MS; supervision, MS; software, XY and MS; validation, LT, LL and MS; visualization, MS; funding, LL and MS; writing—original draft preparation, XY, LT, LL and MS; writing—review and editing, XY, LT, LL and MS.

Correspondence to Long Long or Mohammad Sina.

The authors declare no competing interests.

In improve the language of the manuscript, our sole reliance was on the ChatGPT tool, with the authors asserting full authorship and accountability for its entirety.

All authors are agreeing for participate in this article.

All authors are agreeing for the publication of the manuscript.

Springer Nature remains neutral with regard to jurisdictional claims in published maps and institutional affiliations.

Open Access This article is licensed under a Creative Commons Attribution-NonCommercial-NoDerivatives 4.0 International License, which permits any non-commercial use, sharing, distribution and reproduction in any medium or format, as long as you give appropriate credit to the original author(s) and the source, provide a link to the Creative Commons licence, and indicate if you modified the licensed material. You do not have permission under this licence to share adapted material derived from this article or parts of it. The images or other third party material in this article are included in the article’s Creative Commons licence, unless indicated otherwise in a credit line to the material. If material is not included in the article’s Creative Commons licence and your intended use is not permitted by statutory regulation or exceeds the permitted use, you will need to obtain permission directly from the copyright holder. To view a copy of this licence, visit http://creativecommons.org/licenses/by-nc-nd/4.0/.

Reprints and permissions

Yu, X., Tang, L., Long, L. et al. Comparison of deep and conventional machine learning models for prediction of one supply chain management distribution cost. Sci Rep 14, 24195 (2024). https://doi.org/10.1038/s41598-024-75114-9

Download citation

Received: 08 June 2024

Accepted: 01 October 2024

Published: 15 October 2024

DOI: https://doi.org/10.1038/s41598-024-75114-9

Anyone you share the following link with will be able to read this content:

Sorry, a shareable link is not currently available for this article.

Provided by the Springer Nature SharedIt content-sharing initiative