USING WAVELET TRANSFORM TO IMPROVE QUALITY CLASSIFICATION FOR TIME-SERIES DATA SEQUENCE

Abstract

This paper proposes a solution using wavelet transform to extract features
from a time-series, the outputs of the pre-processing is input of a neural network
in order to classify and predict near future trends of the data. The approach is
based on the CWT and DWT of time-series. The result which is tested on real
datasets HAR (Human Activity Recognition), shows the improvements in
accuracy, reaching 94%. It is an improvement compared to previously reported
results for previous systems.