Accurate and Efficient Mosquito Genus Classification Algorithm Using Candidate- Elimination And Nearest Centroid on Extracted Features of Wingbeat Acoustic Properties

Date of Award

12-2021

Document Type

Dissertation

Degree Name

Doctor of Philosophy in Computer Science

First Advisor

Proceso L. Fernandez, Jr., PhD

Abstract

The automatic identification of mosquito genus from their sounds can be an important technology to help reduce the spread of mosquito-borne diseases. In this study, we explore and develop a simple and yet effective classification technique that outperforms, in terms of accuracy and efficiency, a Convolutional Neural Network (CNN) model in identifying mosquito genus. A data set of sound recordings from the Humbug Zooniverse [7], collected by researchers from Oxford University and actual recordings of mosquitoes in the Philippines were used in this study. Our developed technique involves getting the values from corresponding spectrograms of the audio files, and we initially used three common measures of central tendencies as features -- quartiles, median, and range (minimum and maximum values). Finalizing the model, the quartiles and maximum values produced the highest accuracy scores for the classification. Nearest centroid was then applied to predict the class of an instance by comparing the actual values of the instance against the expected values of each class on these three-mosquito genus. The average classification accuracy of our proposed Descriptive Statistic (DS) model was 97%, and this was higher than the 86% classification accuracy yielded by CNN model. Moreover, the DS model required much less training time and lower classification time than the CNN model, while requiring much less computing memory resources. The results offer a promising technique that may also simplify the process of solving other sound-based classification problems.

Link to Full Text

Share

COinS