| Abstract | The classification model could be used for identifying cancers by gene selection, resource allocation in decision making, ima ge processing in artificial intelligence, and so on. How to classify the unknown samples efficiently and effectively is among the most critical and chall enging procedures in machine learning.
We propose a unified model in feature space based on mathematical programming (UMFS). UMFS a dopts two central ideas. First, it uses a global measurement rather than a local approximation; second, it embeds a multiple criteria and non convex mathematical programming to unify the discriminative learning principles.
Through classification accuracy b ased iterations, UMFS obtains the feature weight vector and finally extracts the optimal feature subset. The performance of t he proposed method is evaluated in extensive experiments on synthetic and real microarray benchmark datasets. Eight classical fe atu re selection methods, four classification models, and two popular embedded learning schemes, including CNN, Fisher's discrimination, k nearest neighbor (KNN), hyperplane k nearest neighbor (HKNN), Support Vector Machine (SVM), Random Forest and Logistic re gression are employed for comparisons. |
