Multiple Classifier Selection to Improve Accuracy of Classifier for Time Series Analysis

摘要

In this article we proposed a technique for temporal data mining which is based on the classification rules and optimal discriminant analysis(ODA).Time series are decomposed into segments (avg,slope,curvature) are described by polynomial models. Then the classifier assesses subsequent segments based on the classification rule activity. And assign an input a class. Segmentation and piecewise polynomial modeling are done fast over time series. For this classifier we use Euclidean distance measure for time series and using a fast Fourier Transform (FFT) to construct a multiple dynamic classifier to increase the accuracy of a classifier.