Analysis of various classification and clustering algorithms for identification of flying insects

Abstract

This research work aims in identifying insects, then recognizing insects and finally evaluating the vector density of insects based on their flight sound using Modified Support Vector Machine with Tukey s method and Hidden to Observable Markov Model (H2O-MM) with Binary Matrix Shuffling Filter (BMSF) optimization techniques. Entomologists nowadays are now in need of an economical tool in identifying the insects accurately. Flight sound of insects are characterized by many features, especially, this research work focuses on spectral and temporal features. A complete digitized tool is developed for detecting and monitoring insects threatening biological resources. This research work also implemented framework that compares several classification and clustering algorithms with same data sets. The methodology uses Turkey s method to optimize the feature selection and modifies Support Vector Machine (SVM) classifier in order to detect the insects. The comparative study is performed on Benchmark data sets MOSQUITO, ESC-50 and a synthetic dataset as uploaded on Kaggle dataset using the Bayesian classifier, k-Nearest Neighbor, Fisher Linear Discriminant Analysis Algorithm, Neural Network classifier, and Fuzzy classifier against Support Vector Machine classifier. Superficially Modified SVM outperforms other classifiers with accuracy 86.76%. The aims is to recognize the type of insects using a novel Hidden-to-Observable Markov Model clustering optimally selected features, using BMSF for optimized featureselection that lead to good accuracy of recognition of type of insects. Additionally, it approximately estimates the vector density, ie. number of insects based on the cluster density. Extensive experiments are conducted on the Benchmark dataset, ESC-50 and the Davies Bouldin (DB) Index is estimated to 0.46 with BMSF on H2O-MM clustering. A comparative study is done on various clustering algorithms of statistical nature (Agglomerative Hierarchical clustering (AHC), k-means, Expected Maximization (EM)) and soft computing