Time series classification by similarity and extraction of intrinsic attributes with application to insect species identification

Abstract

Integrating sequential and temporal data into the Data Mining process is of one of the most important challenges in Machine Learning, for tasks such as clustering, classification and prediction. In the case of classification, there exist two approaches frequently used with temporal data: similarity and feature extraction. Classification by similarity uses a distance function to identify the most similar time series to a query series, and the query class is assigned to the dominant class among the similar time series; the classification by feature extraction performs a search for local features and uses those features to induce a classifier. In this project we are interested in classifying signals from disease vectors mosquitoes using time series data obtained from an optical sensor. Our objective is to compare classification methods by similarity and feature extraction applied to this context. Our previous experience indicates that the feature extraction approach has shown the most promising results. Given the large volume of insect data collected so far, our main object is to research, evaluate and compare automatic and non-supervised methods for intrinsic feature identification. Some of these methods have been widely used by the signal processing community to classify several types of sound signals, such as musical notes.