A new unsupervised pre-processing algorithm based on artificial immune system for ERP assessment in a P300-based GKT

• Main Authors: Shojaeilangari, Seyedehsamaneh, Moradi, Mohammad Hassan
• Other Authors: School of Electrical and Electronic Engineering
• Format: Article
• Language: English
• Published: 2015
• Subjects: DRNTU::Science
• Online Access: https://hdl.handle.net/10356/103205; http://hdl.handle.net/10220/25770; http://maxwellsci.com/jp/abstract.php?jid=RJASET&no=214&abs=07
• Institution: Nanyang Technological University

In recent years, an increasing number of researches have been focused on bio-inspired algorithms to solve the elaborate engineering problems. Artificial Immune System (AIS) is an artificial intelligence technique which has potential of solving problems in various fields. The immune system, due to self-regulating nature, has been an inspiration source of unsupervised learning methods for pattern recognition task. The purpose of this study is to apply the AIS to pre-process the lie-detection dataset to promote the recognition of guilty and innocent subjects. A new Unsupervised AIS (UAIS) was proposed in this study as a pre-processing method before classification. Then, we applied three different classifiers on pre-processed data for Event Related Potential (ERP) assessment in a P300-based Guilty Knowledge Test (GKT). Experiment results showed that UAIS is a successful pre-processing method which is able to improve the classification rate. In our experiments, we observed that the classification accuracies for three different classifiers: K-Nearest Neighbourhood (KNN), Support Vector Machine (SVM) and Linear Discriminant Analysis (LDA) were increased after applying UAIS pre-processing. Using of scattering criterion to assessment the features before and after pre-processing proved that our proposed method was able to perform data mapping from a primary feature space to a new area where the data separability was improved significantly.