Comprehensive Machine Learning Approaches for Predicting and Analysing Abnormal Behaviour Patterns
A Supriya, Assistant professor, Anantha Lakshmi institute of Technology and sciences, Anantapur
H Prasanth Kumar, Assistant professor, Anantha Lakshmi institute of Technology and sciences, Anantapur
Abstract
In specific environments, improper behaviours such as smoking at a gas station can lead to significant safety hazards, making the timely and accurate detection of such actions crucial. This paper investigates the application of machine learning algorithms to predict and identify abnormal behaviours effectively. A comprehensive dataset comprising 1,200 samples was collected, categorized into three behaviour classes: smoking (30%), making phone calls (35%), and normal activities (35%). To evaluate predictive performance, six widely recognized machine learning algorithms were implemented: Linear Support Vector Machine (LSVM), Kernel Support Vector Machine (KSVM), Decision Tree Classifier (DT), Random Forest Classifier (RF), K-Nearest Neighbours (KNN), and K-Means Clustering.
The performance of these algorithms was assessed using various metrics, including accuracy, precision, recall, F1-score, and Mean Squared Error (MSE). Among the tested models, the Random Forest Classifier (RF) emerged as the best performer, achieving an overall accuracy of 82%, with a precision of 84%, recall of 80%, F1-score of 82%, and an MSE of 0.18. Comparative analysis revealed that the Random Forest Classifier outperformed other algorithms due to its robustness in handling complex feature interactions and imbalanced datasets.
Additionally, Principal Component Analysis (PCA) was employed to visualize the classification results, demonstrating clear separation between the behaviour categories. This visualization further validated the model's predictive capability. The findings of this study indicate that Random Forest Classifier provides a reliable and efficient approach for predicting abnormal behaviours, offering potential applications in safety-critical scenarios such as industrial workplaces, public spaces, and monitoring systems. Future work aims to enhance the model’s accuracy through larger datasets, real-time prediction capabilities, and integration with advanced feature engineering techniques.
