CoCoST: A Computational Cost Efficient Classifier

摘要: Computational cost of classification is as important accuracy in on-line systems. The computational usually dominated by the computing implicit features raw input data. Very few efforts have been made to design classifiers which perform effectively with limited power; instead, feature selection employed a pre-processing step reduce running traditional classifiers. We present CoCoST, novel and effective approach for building achieve state-of-the-art accuracy, while keeping expected low, even without selection. CoCost employs wide range cost-aware decision trees, each tuned specialize classifying instances from subset space, judiciously consults them depending on instance accordance meta-classifier. Experimental results network flow detection application show that, our can better than such SVM random forests, achieving 75%-90% reduction costs.