A Method to Extract Feature Variables Contributed in Nonlinear Machine Learning Prediction.

摘要: Background Although advances in prediction accuracy have been made with new machine learning methods, such as support vector machines and deep neural networks, these methods make nonlinear models thus lack the ability to explain basis of their predictions. Improving explanatory capabilities would increase reliability Objective Our objective was develop a factor analysis technique that enables presentation feature variables used making predictions, even models. Methods A consisted two techniques: backward extraction technique. We developed extracted obtained from posterior probability distribution model which calculated by Results In evaluation, using gene expression data prostate tumor patients healthy subjects, networks approximately 5% better than machines. Then rate concordance between an earlier report Jensen–Shannon divergence ones this elimination Hilbert–Schmidt independence criteria 40% for top five variables, 10, 49% 100. Conclusion The results showed can be evaluated different viewpoints techniques. In future, we hope use verify characteristics features technique, perform clinical studies genes, experiment.