Multi - Model Fusion Learning Method And Its Application

Machine learning aims at improving generalization automatically on empirical data. It has become the main method of intelligent data analysis. With the rapid increase of business demands, machine learning has been used to predict the needs of users and other applications. A simple and effective way to achieve better performance is to combine multiple learners. This paper applies multiple-learner method to brand recommendation, stock recommendation, and robot landmark exploration, described in the following:For the problem of recommending brands in online shopping malls, we propose a framework based on combining multiple learners. This framework uses many ways to enhance the diversity of individual learners, thereby enhancing the performance after combining. This framework achieved good result on the Alibaba big data competition.For the problem of recommending stocks, we propose a method by combining double average stock selection model, machine learning model based on traditional stock data, and machine learning model based on new media stock data. This method achieved higher reward than the benchmarks significantly in a certain period of time.For the problem of robot landmark exploration, we propose a landmark exploration strategy which combines the global location estimation and Kalman filter. This strategy can effectively rectify past errors of the Kalman filter when the robot has a large angle direction turning. We demonstrate the effectiveness of this strategy in a simulated environment.