Optimized pattern matching in time series data using segmentation techniques

Abstract

In recent years, time series data mining has become an active research area due to the rapid proliferation of temporal-dependent applications. Owing to the continuous updation of data streams at high bit rates, storing and processing the entire data streams becomes impractical. Hence, it is essential to reduce the dimensionality. Pattern matching in time series often faces significant challenges due to the inconsistent and incomplete data. Most of the dimensionality reduction schemes are designed based on the assumption that every class of samples follows the Gaussian distribution. Lack of this property in real time data distribution does not allow dimensionality reduction techniques to characterize the different classes well and measure the data uncertainty accurately. This research work addresses these constraints in the pattern matching model through three contributions such as Pattern Matching using Ant Colony Optimization (PM-ACO), Pattern Matching using Particle Swarm Optimization (PM-PSO), and Handling UNcertainty and missing value prediction in Time series (HUNT). The initial work is PM-ACO that employs two segmentation methods such as Perceptually Important Points Method (PIP) and Piecewise Aggregate Approximation Method (PAA). It applies the segmentation method as the preprocessing step of the pattern matching process. Then, PM-ACO stores the segmented patterns in an Optimal Binary Search Tree (OBST) based on either time series characteristics or mean values. By applying the ACO, it matches the current window of the time series data with a specific pattern efficiently. The PIP based proposed pattern matching approach has gained 18% precision, 20% recall, 18.7% F-measure, and 25% accuracy than the existing Brute force based pattern matching approach. Compared to the Brute force based pattern matching approach, the PAA segmentation based PM-ACO approach has improved the precision by 20%, recall by 17%, Fmeasure by 18.4%, and accuracy by 15%. newline