Intrusion Detection and Classification Using Machine Learning for Manufacturing Surveillance Systems

The growing number of security threats in manufacturing and industrial settings has revealed the shortcomings of conventional surveillance systems, which are primarily based on manual surveillance and basic rule-based techniques. These traditional methods commonly tend to miss suspicious behavior in real time, resulting in delayed reactions and a greater risk of illegal access or criminal activities. With the introduction of sophisticated machine learning methods and access to dense surveillance datasets, there is increased potential to design smart systems that can automatically detect and classify intrusions. The present work intends to solve the problem of manufacturing surveillance improvement through the use of Machine Learning (ML) models—namely Multilayer Perceptron (MLP), XG-Boost, and Random Forest (RF) to design a precise and effective intrusion detection and classification system. Through the UCF-Crime dataset, which offers varied and realistic surveillance video clips, the research centers on preprocessing the data, training and testing the chosen models, and comparing the performance of the models based on standard parameters like accuracy, precision, F1-score, and recall MLP classifier was superior to the other models, with an accuracy of 99.44%, a precision of 99.45%, a recall of 99.40%, and an F1-score of 99.42%. The XG-Boost model was followed by an accuracy of 95.7%, a precision of 96.33%, a recall of 94.59%, and a F1-score of 95.39%. The RF model, being good, had relatively lower values, with an accuracy of 98.1%, precision of 97.56%, recall of 97.5%, and F1-score of 97.57%.