T-EKC: A Novel Ensemble Learning for Classification Application

Data classification plays a key role in the research area of reducing the rate of patient misdiagnosis and improving the accuracy of equipment fault detection. However, existing most classification methods only pinpoint binary class data, which is unreasonable for practical applications. As a significant machine learning method, ensemble learning addresses classification problems with multiple basic classifiers. In this paper, T-EKC, a new classification method based on ensemble learning is proposed for binary class, multivariate class and imbalanced class datasets. ELM(extreme learning machine), KNN(K nearest neighbors) and CART(classification and regression tree) are employed to obtain initial classification results. The results of each classifier are then combined by EWVE(Entropy weight voting evaluation) to establish the final output. Furthermore, parameter perturbation is developed for the diversity problem of ensemble learning and the Friedman test is adopted to implement the difference significance test of the proposed T-EKC method. A large number of experiments are implemented on 15 different types of datasets, and experimental results indicate that T-KEC is effective and efficient in data classification with classification accuracy of 85.67% on average for different types of datasets.