Selection of statistical wavelet features using a wrapper approach for electrical appliances identification based on a KNN classifier combined with voting rules method

Abstract

This work is an extended version of a paper presented in the International Conference on Intelligent Systems and Patterns Recognition where the authors have proposed a compact features representation based on the estimation of statistical features using discrete wavelet transform for electrical appliances identification based on a K nearest neighbour classifier combined with voting rule strategy. The results have shown that the wavelet cepstral coefficients (WCC) descriptor presents highest performance with 98.13% classification rate (CR). In this work, we propose many extensions: 1) The logarithm energy (LOG_E) is used as additional descriptor; 2) The relevance of the wavelet-based features combined with LOG_E descriptor is investigated using feature selection based on wrappers approach; 3) Deep performances evaluation is carried out using five additional metrics. The results show that the selection of four features of WCC combined with LOG_E improves the CR at 98.51%.