Identifying relevant features of termite signals applied in termite detection system

Abstract

The termite detection system has been studied intensively in both industry and academia. However, the devices currently in the market are only used to detect the presence of termites. Therefore, this research aims to develop a termite detection system capable of detecting the presence of termites and predicting their population size. This research focuses on the extraction and the relevant feature selection processes on acoustic and temperature signals, which complement the new system's design. To this end, the Boruta package was employed to identify relevant feature sets in the signals that can be used to significantly distinguish the different termite population sizes. The sample consists of 40 acoustic and 10 temperature features extracted and integrated with the Boruta package. However, after proper numerical analysis, an identified total of 25 relevant features consisting of 21 acoustic and 4 temperature features were used. Therefore, by reducing the dimensionality of the data set with the Boruta package, the computational burden of the termite detection system can be decreased. This study confirms that implementing relevant features to the termite detection system provides better performance due to its detection accuracy of 97.167% and ability to predict termites population size with a root mean squared error (RMSE) of 98.316.