Sequence Mining-Based Support Vector Machine with Decision Tree Approach for Efficient Time Series Data Classification

Abstract

The growing demand for an efficient approach to classify time series data is bringing forth numerous research efforts in data mining field. Popularly known applications like business, medical and meteorology and so on, typically involves majority of data type in the form of time series. Hence, it is crucial to identify and scope out the potential of time series data owing to its importance on understanding the past trend as well as predicting about what would occur in future. To efficiently analyze the time series data, a system design based on Sliding Window Technique-Improved Association Rule Mining (SWT-IARM) with Enhanced Support Vector Machine (ESVM) has been largely adopted in the recent past. However, it does not provide a high accuracy for larger size of the dataset along with huge number of attributes. To solve this problem the proposed system designed a Sequence Mining algorithm-based Support Vector Machine with Decision Tree algorithm (SM-SVM with DT) for efficient time series analysis. In this proposed work, the larger size of the dataset is considered along with huge number of attributes. The preprocessing is performed using Kalman filtering. The hybrid segmentation method is proposed by combining a clustering technique and Particle Swarm Optimization (PSO) algorithm. Based on the sequence mining algorithm, the rule discovery is performed to reduce the computational complexity prominently by extracting the most frequent and important rules. In order to provide better time series classification results, the Support Vector Machine with Decision Tree (SVM-DT) method is utilized. Finally, the Pattern matching-based modified Spearmen’s rank correlation coefficient technique is introduced to provide more similarity and classification results for the given larger time series dataset accurately. The experimental results shows that the proposed system achieves better accuracy, time complexity and rule discovery compared with the existing system.