Entropy based Probabilistic SVM Learning for Complex Processes

The processes in biomedical engineering, biology and social system are usually much more complex than machinery processes because of high-dimensional nature, limited data available and stochastic uncertainties. Data based learning becomes an only solution for decision making due to little available knowledge of the process. It is always a great challenge to modeling and learning of the high-dimensional process under small data samples and strong uncertainties.A novel entropy-based probabilistic methodology is proposed to improve the shortcoming of the traditional data classification. The entropy method is used to reduce the model complexity. A probabilistic SVM is constructed in three different components: similarity based sampling, distributed SVM systems, and probability-based optimization. Using this method, the data uncertainties can be properly extracted. The influence of the data uncertainty to the decision boundary can be considered in the objective function. Through the probabilistic optimization, a reliable classification will be obtained.