CFS-DT : a Combined Feature Selection and Decision Tree based Method for Octane Number Prediction

Abstract

Octane number (ON) is the most important index of vehicle gasoline specification. Due to the complexity of refining process, the equipment variety, a large number of features are collected, which makes it difficult to predict ON of gasoline. In this paper, we propose a combined feature selection and decision tree based prediction method, CFS-DT, which combines low variance filtering, high correlation filtering and random forest to execute feature selection on a large number of original feature first. After that, a decision tree(DT) is trained for ON prediction on selected features. Experiments are carried out on datasets collected from 2020 Huawei cup Mathematical Modeling show that our model has a good effectiveness and achieves a 89% prediction precision.