Design and Implementation of an Intelligent Business to Business Stock Control System Using Machine Learning Technique

Abstract

Stock control is a requirement for automatic monitoring of merchandise within a stock database and predicts when it is due for restocking in real-time. To achieve this the methodology used are data collection, data analysis, feature selection, feature transformation, machine learning, and the Internet of Things (IoT) to develop a stock control system. The data used is a 3-year Out of Stock(OOS) records in Shoprite Enugu State shopping mall, was collected and subjected to a series of processing steps. The processed data features were selected with Chi-square and then transformed into a compact feature vector using Principal Component Analysis (PCA) to train a Feed-Forward Neural Network (FFNN) algorithm and generate a model for OOS prediction. Upon achieving this, an IoT algorithm that utilized Simple Mail Transfer Protocol (SMTP) was used to notify the stock admin of the need for restocking of the identified OOS product. The system was implemented using MATLAB and JavaScript programming language. The results of the evaluation process showed that the proposed model recorded tolerable error during the training process with a Mean Square Error (MSE) of 0.17168 and a Regression (R) of 0.96907, which suggested a very good prediction model. To validate the model, a k-fold cross-validation approach was applied, and the results recorded an MSE average of 0.09686, while the R reported 0.97168. Comparative analysis with other state-of-the art algorithms was performed, considering the MSE results of the new and existing OOS prediction models, and the results showed that the new model was among the best three performing models compared. However, the new model, due to its IoT features, was the most reliable as it was capable of notifying the stock admin in real time of the stock status and the need for restocking of products. Experimental validation of the model as a software considering several products which are running out of stock showed the ability of the system to monitor in real time and notify the admin through email on the need to restock the products