Cutting-Edge Patterns of Preemption: Classifying Violent Incidents with CNN-SVM Intelligence

This research paper presents a comprehensive analysis of violence detection using a hybrid CNN-SVM approach across five distinct violence classes: for instance, gangs, physical assault, domestic violence, bullying, rioting, and verbal abuse. The study presents a neural network of the convolutional type that is aimed to use deep learning in feature extraction from data visualization and, next, an SVM that is meant to do classifications. Performance metrics for each class reveal the nuanced capabilities of the model: Class 1 showed precision, recall, and F1-score of 77.83%, 75.84%, and 76.82 % respectively; Class 2 exhibited higher recall with values of 78.21 % precision, 86.17% recall, and an F1-score of 82.00%; Class 3 maintained a balance with an 80.43 % precision, 78.90% recall, and a 79.66% F1-score; Class 4 demonstrated vital precision and recall at 80.54% and 85.18% respectively, with an F1-score of 82.80%; Class 5 achieved the highest precision of 88.59%, with a recall of 79.82% and an F1-score of 83.98%. The overall model performance is captured using various averaging techniques: Three different ways of computing values were explored through macro average (of 81.12% precision, 81.18% recall, and 81.05% F1-score), weighted average (of 81.21 % precision, 81.01% recall, and 81.01 % F1-score); and micro average (that shows uniform values throughout of Such outcomes clearly show significant progress of a CNN-SVM model in classifying among complexly occurring violence patterns that can be helpful for an implementing of automatic violence detection systems with substantive application. Research presented a different perspective on class classification performance that was quite sensitive but had relatively high efficiency, as shown by the precision of 90% -94% among the five classes.