Fuzzy c-Means Classifier with Deterministic Initialization and Missing Value Imputation

A fuzzy c-means (FCM) classifier derived from a generalized FCM clustering is proposed. The classifier design is based on FCM. The classifier is not initialized with random numbers, hence being deterministic. The parameters are optimized by cross validation (CV) protocol and golden section search method. A method for dealing with missing values without eliminating them but with estimating them is also proposed. Instead of using the terminology "conditional expectation", the imputation is done by the least square method of Mahalanobis distances between the datum with missing values and cluster centers. The FCM classifier outperforms well established methods such as support vector machine, k-nearest neighbor and Gaussian mixture classifiers for datasets with and without missing values