Score Prediction Based On NN-Stacking

Abstract:
Performance prediction is a research hotspot of education data. At present, the most widely used machine learning algorithms include neural networks, decision trees, SVM, linear regression, and Bayesian networks. When using a single machine learning method, you often get a model with a certain preference in a certain aspect, and the performance of this model is usually not stable enough. Ensemble learning is also a kind of machine learning technology. It combines multiple base classifiers and integrates these classifiers through certain strategies. Compared with a single learner, the final model has better generalization ability and performance and more stable. The Stacking ensemble method uses linear combination to fuse multiple regression models to improve the prediction accuracy. This paper uses NN-Stacking to improve the linear stacking algorithm proposed by Breiman by allowing linear parameters to vary with the input characteristics. The improved NN-Stacking algorithm uses neural network to predict the superposition coefficient, which can make different basic models perform better in different areas of the feature space, and maintain the same interpretation characteristics as the linear stacking algorithm at the local level, which improves the accuracy of performance prediction sex.