STDPboost: A Self-Training Method Based on Density Peaks and Improved Adaboost for Semi-Supervised Classification

The self-training methods have been praised by extensive research in semi-supervised classification. Mislabeling is the main challenge methods. Multiple variations of are recently proposed against mislabeling from following one two aspects: a) using heuristic rules to find high-confidence unlabeled samples that can easily be predicted correctly each iteration; b) enhancing prediction performance employing ensemble classifiers composed multiple weak classifiers. Yet, they still suffer issues: a); most strategies for finding heavily rely on parameters; almost all employed originally designed supervised and may not suitable classification due limited number unrepresented distribution initial labeled data; c) few overcome above aspects at same time. To advance state art, a new method based density peaks clustering improved Adaboost presented named as STDPboost. In iterative self-taught process, clustering-based strategy classifier AdaboostSEMI more predict samples, which overcomes mentioned shortcomings existing Intensive experiments benchmark data sets proven STDPboost outperforms 7 state-of-the-art average accuracy KNN CART with percentages 10% 50% further alleviating mislabeling.