Fuzzy kNN Adaptation to Learning by Example in Activity Recognition Modeling

Abstract

Activity Recognition is a complex task of the Human Computer Interaction (HCI) domain. k-Nearest Neighbors (kNN) a non-parametric classifier, mimics human decision making, using experiences for segregating a new object. Fuzzy Logic mimics human intelligence to make decisions; but suffers from requiring domain expertise to propose novel rules. In this paper a novel technique is proposed that comes with efficient fuzzy rules from the training data. The kNN classifier is modified by incorporating fuzzification of the feature space by learning from the data and not relying solely on domain experts to draw fuzzy rules. Additional novelty is the efficient use of the Fuzzy Similarity Relations and Fuzzy Implicators for hybridization of the kNN Classifier. The proposed hybridized fuzzy kNN classifier is shown to perform 5.6 % better than the classical kNN counterpart.