Time series data analysis using Artificial Immune System

Article type: Research Article

Authors: Ahmad, Waseem^{a; *} | Narayanan, Ajit^b

Affiliations: [a] Toi Ohomai Institute of Technology, New Zealand | [b] AUT University, New Zealand

Correspondence: [*] Corresponding author: Waseem Ahmad, Toi Ohomai Institute of Technology, New Zealand. E-mail: [email protected].

Abstract: Extracting knowledge from temporal data has become an important topic of research for data mining researchers due to its wide range of applications in real world problems, such as finding patterns in weather data for forecasting, stock market data for prediction and industrial production indices forecasting. In this paper, a novel time series data analysis approach inspired by the processes of natural immune system is proposed. The approach is layered where, firstly, segmentation is used to sub-divide the complete time series data into sub-sequences and, secondly, an Artificial Immune System (AIS) algorithm is used to analyse and cluster segmented data. Finally, this clustering information is used to build a model for forecasting and prediction. The effectiveness of the proposed approach is demonstrated by testing it on various non-linear cyclic time series data and results are compared against linear regression and multi-layer perceptrons.

Keywords: Temporal data mining, Artificial Immune System, data segmentation, clustering, B-cells, antibodies

DOI: 10.3233/IDT-170315

Journal: Intelligent Decision Technologies, vol. 12, no. 2, pp. 119-135, 2018