抄録
Fuzzy c-means-based classifier derived from a generalized
fuzzy c-means (FCM) partition and tuned by particle
swarm optimization (PSO) has been proposed. Since different
types of classifiers work best for different types of data, our approach is to parameterize the classifiers and tailor them to individual data set. The procedure consists of two phases. The first phase is an unsupervised clustering, which is not initialized with random numbers, hence being deterministic. The second phase is a supervised classification. The parameters of membership functions are optimized by the PSO and cross validation (CV) procedures. The FCM classifier has following advantages. 1) Classification performance in terms of 10-fold CV and three-way data splits (3-WDS) is high. 2) Missing values can be estimated based on the least square Mahalanobis distances. 3) High-dimensional feature
vectors can be classified taking into account the covariance structure of clusters. 4) The relational version can be used when many of the feature values are missing, or when only relational data are available instead of the object data.
