Gunshot Audio Analysis for Firearm Type Detection Using Machine Learning Algorithms
Loading...
Files
Date
Authors
Journal Title
Journal ISSN
Volume Title
Publisher
Universiti Malaysia Sarawak (UNIMAS)
Abstract
Description
Current gunshot detection systems primarily focus on identifying the presence of gunfire but lack the ability to classify the specific type of firearm. This thesis presents a machine learning-based system designed to classify firearm types based on gunshot audio recordings. A dataset consisting of 851 gunshot recordings across eight firearm classes was used. 108-dimensional feature vector comprising Mel-frequency cepstral coefficients (MFCCs), Chroma, Spectral Contrast, Zero-Crossing Rate (ZCR), Energy, and Spectral Bandwidth. The features were normalized and used to train three classification models: Support Vector Machine (SVM), k-Nearest Neighbors (kNN), and Convolutional Neural Network (CNN). Each model was optimized using grid search with 5-fold stratified cross-validation. Experimental results showed that the SVM achieved an accuracy of 93.26%, while both the kNN and CNN models achieved higher accuracies of 96.77%. A web-based application was developed using Django, allowing users to upload gunshot audio and select a preferred model for real-time firearm classification. Although the system performs effectively on the curated dataset, its accuracy decreases when applied to real-world audio samples, indicating the need for improved generalizability. This project demonstrates the potential of machine learning for firearm identification from audio signals and provides a functional prototype contributing to intelligent surveillance and public safety technologies.