Published Paper Details:

Anomaly detection is a critical task with applications spanning various domains, including manufacturing, healthcare, and security and surveillance. This study introduces a comprehensive deep learning-based anomaly detection framework for video data using PyTorch. The architecture, built upon the "FNN" (Feedforward Neural Network) is "AnomalyDetectNet", employs multiple layers for feature extraction and classification, incorporating innovative techniques like Multiple Instance Learning (MIL) and the hyperbolic tangent (tanh) activation function to distinguish between regular and anomalous data effectively. Evaluation of benchmark datasets, particularly the UCF-Crime dataset, through metrics like the Area Under the Receiver Operating Characteristic Curve (AUC-ROC), demonstrates the model's proficiency in anomaly detection. Comparative analysis against a baseline approach reveals notable improvements, with our "AnomalyDetectNet" model achieving an accuracy of 84.84%, surpassing the baseline's 84.30%. This work contributes a potent tool for real-world applications in surveillance, security, and anomaly detection across diverse scenarios, advancing the field of anomaly detection research.