Identification of human gait using genetic algorithm tuned fuzzy logic

Data mining is concerned with the discovery of useful hidden information in large databases. Classification is a data mining task producing rules in which a set of attributes in data predict the value of a class attribute. Classifiers usually produce a large number of rules, most of which are not interesting to the user. Rule interestingness is a decisive factor. However, evaluating rule interestingness is challenging as it involves both objective (data-driven) and subjective (user-driven) aspects.;In this research, a fuzzy genetic algorithm is proposed to discover classification rules that are both accurate and interesting. Continuous attributes are fuzzified so that the produced rules are fuzzy rules stated in terms that are more natural to users and easier to measure. A weighted fitness function is used with two elements: the first is an objective interestingness measure based on the attribute information gain, and the second is a predictive accuracy measure.;Classification of human gait dynamics data is useful for rehabilitation processes. Three variants of the proposed fuzzy genetic algorithm are experimented for different classification tasks performed on a collected gait dynamics dataset from 23 participating healthy subjects. Part of the dataset is used for training of the genetic algorithm and the other part is used to test the performance of the genetic algorithm. Promising results are obtained specially for identifying the human subjects based on their gait dynamics, and mapping an unknown subject to a previously known subject with similar gait parameters.