Machine Learning Analysis on Gunshot Recognition

The ability to recognize a gunshot has significance in reinforcing public safety, assisting in crime scene investigations, and preventing gun violence. This paper investigates the efficiency of various machine learning models for gunshot recognition. We present a model to identify the type of pistol or rifle discharged by analyzing only an audio signal of the gunshot. Among the array of methods explored, AdaBoost performed the best achieving an accuracy of 99.9% and sustaining over 80% accuracy with 40 dB conditions. Additionally, we experimented with the importance level of the features to identify the most relevant variables that boost the performance of the algorithms.