AN EFFECTIVE PREDICTION SYSTEM FOR TIME SERIES DATA USING PATTERN MATCHING ALGORITHMS

The need for forecasting has increased significantly due to the rapid changes in technology, social changes and for making the future plans. The main objective of the paper is to improve the accuracy of the prediction algorithm for time series data. To perform the prediction, the dataset is first clustered and then labeling is provided for samples in the dataset. For clustering, existing research work uses K-Means (KM) clustering algorithm despite the fact that, this algorithm is more sensitive to initialization. To counter this limitation, K-Harmonic Means (KHM) algorithm is used in this research work for data clustering. Such clustering approaches are used for finding pattern matching sequences in the samples. For predicting the time series data, this research work proposed Pattern Sequence-based Prediction Algorithm along with Weighted Moving Average Method (PSPA-WMAM). PSPA is used to extract the values next to the matched sequences, and then Weighted Moving Average Method (WMAM) is applied to the extracted values for final prediction. The advantage of WMAM is that more weight values are allocated to the recent data, and less weight values are allocated to the old data. For demonstrating the use of the proposed algorithm two different datasets: Australian Electricity dataset and Tamilnadu Weather dataset are considered in this work. The prediction results obtained based on the KM and the KHM clustering are compared using statistical measures for observation and interpretation of the significance of the proposed approach. Empirical results on datasets showed that the proposed Pattern Sequence-based Prediction Algorithm along with Weighted Moving Average Method (PSPA-WMAM) statistically outperforms, and hence demonstrates that the proposed approach is an effective algorithm for predicting the time series data.