Activity recognition using optimized reduced kernel extreme learning machine (OPT-RKELM) / Yang Dong Rui

Yang , Dong Rui (2019) Activity recognition using optimized reduced kernel extreme learning machine (OPT-RKELM) / Yang Dong Rui. Masters thesis, Universiti Malaya.

	PDF (The Candidate's Agreement) Restricted to Repository staff only Download (179Kb)
Preview	PDF (Thesis M.A) Download (884Kb) | Preview

Abstract

In the past decade, research related to Human Activity Recognition (HAR) based on devices embedded sensors has shown good overall recognition performance. As a consequence, HAR has been identified as a potential topic for healthcare assessment systems. One of the major research problems is the computation resources required by machine learning algorithm used for classification for HAR. Numerous researchers have tried different methods to enhance the algorithm to improve performance, some of these methods include Support Vector Machine (SVM), Decision Trees, Extreme Learning Machine (ELM), Kernel Extreme Learning Machine (KELM), and Deng’s Reduced Kernel Extreme Learning Machine (RKELM). However, unsatisfactory accuracy, slow learning speed, and stability is still a problem. In this study, we have purposed a model named as Optimized Reduced Kernel Extreme Learning Machine (Opt RKELM). It applies the characteristic of ReliefF algorithm to rank and select top scoring features for feature selection. ReliefF can solve the problem of large feature dimension in the existing RKELM. By using clustering method K-Means, we have found the best center point position to calculate Kernel matrix. at last, we have employed Quantum-behaved Particle Swarm Optimization (QPSO) to get the optimal kernel parameter in the proposed model. To evaluate the effectiveness of Opt-RKELM, two benchmark datasets related to human activity recognition problems are used. The notable advantages of the proposed model are excellent recognition accuracy, fast learning speed, stable prediction ability, and good generalization ability.

Actions (For repository staff only : Login required)