A Comparative Study Using Two Classifiers For Hazardous Audio Event Classification

Abstract

The hazardous acoustic event classification system is presented and tested in threatening environments. The system is based on classified with Support Vector Machine (SVM), k Nearest Neighbor (kNN) and modeled with Genetic Regulatory Network (GRN). GRN is adopted as classification framework and greatly reduced input feature dimensions. Setting the results that have already reduced the inputs dimensions from GRN framework as inputs for SVM and kNN can correctly classify audio event with low computational time and cost. Comparative and classification tests are carried out using three kinds of input sets with SVM and kNN classifier. These input sets are original feature set, reduced dimension feature set by GRN and unique feature set. SVM applies as novel discriminative approach for dissimilarity measure in order to address a supervised sound-classification task and then shows good performance in the task of acoustic event classification. Selecting GRN in event classification system can not only reduces cost and effort but also aims to obtain high performance and accuracy in varying nature of environments.