A Reinforcement-Algorithm Framework for Automatic Model Selection

Abstract

In recent years, with the development of artificial intelligence, how to realize the automation of machine learning has become a very popular research item. Among them, automatic model selection is an important part. For traditional machine learning, the part of the feature preprocessing, feature selection, model selection and so on are all needed manual processing, and they will take too much time and human resource, so automatic model selection is proposed to reduce resource waste. The automatic model selectionis based on the feature distribution of the data, and the model can be trained to obtain a model with better performance. This paper proposes an automatic model selection framework based on reinforcement learning. The framework can be divided into two phases, one is data preprocessing stage, which uses meta-learning to process; the other is model se lection framework, including feature preprocessing, feature selection and model selection. These three aspects will be the environment of reinforcement learning, through which a model with the highest accuracy can be chosen as the predictive model output. It has been proved by experiments that the method is effective.