SSL-EDA: semi-supervised learning algorithm based on estimation of distribution algorithm
Jian Cong Fan, Yong Quan Liang · International Journal of Innovative Computing and Applications · 2012
Estimation of distribution algorithm (EDA) is a new branch of evolutionary algorithms. EDA replaces search operators with the estimation of the distribution of selected individuals + sampling from this distribution. A semi-supervised learning algorithm based on EDA (abbr. SSL-EDA) is designed. SSL-EDA uses a few data samples with class labels to estimate class distributions of a mount of data instances without class labels. Each data is an individual and the initial labelled individuals are treated as initial population. The optimum individuals can be obtained from the probabilistic distributions of former generation. The local classification rules are produced according to the properties of the optimum individuals. New individuals without labels are selected according to the local classification rules and added with labels to compose new population combined with the optimum individuals. SSL-EDA is compared with several classification algorithms in error rates of classification and also with standard genetic algorithms. The experimental and analytical results show SSL-EDA is better than or comparable with other algorithms in classification accuracy.