The Applications Of Feature Selection Based Classification Ensemble In The Spleen Feebleness Diagnosis

Data mining draws much attention in the computer-aided diagnostic system. In fact, computer-aided diagnosis is a classification task. The clinical data in Chinese medicine, which distinguishes from the data in other fields, is incomplete and ambiguous. These features make data mining in Chinese medicine much difficult.Most of researchers hope to find a "best" classifier to minimize error rate. However, there is no "best" classifier. In fact, the amount of error rate minimized by individual classifier is limited due to the small amount of clinical data. This paper tries to combine several classifiers, which have relatively low error rate and diversity.Research has shown that, to be an effective classification ensemble algorithm, individual classifiers must have relatively low error rate and diversity between them. The method of disarranging training data is often applied to obtain high diversity between individual classifiers by general classification ensemble algorithms, such as Boosting and Bagging. This paper proposes a new classification ensemble model which is based on feature selection. This model applies the method of disarranging both training data and input attributes to obtain individual classifiers with lower error rate and higher diversity between them. Comparison experiments show that the new model is able to lead lower error rate than individual classifiers and other ensemble learning algorithms in the Spleen Feebleness diagnosis. Further experiments indicate that obviously positive effect in minimizing error rate is reached due to the effort of feature selection.