Smart learning has been concerned in the domain of education, and how to correctly predict students' performance is one of meaningful issues. Mature performance prediction methods are promising to be applied to ed...
详细信息
Smart learning has been concerned in the domain of education, and how to correctly predict students' performance is one of meaningful issues. Mature performance prediction methods are promising to be applied to education systems to make educators know more about students' learning states and provide students with academic help in time. Based on students' historical course results and courses' basic information, our work proposes a student's course results prediction model using reasonable dataprocessing and machine learning methods, which can well achieve the goal of out-of-sample results prediction. Our model framework mainly has two parts: numeric and non-numeric features vector embedding algorithms;model optimization based on data augmentation and integration. Firstly, we generate respectively feature embedding vectors for non-numeric and numeric data;Secondly by applying relative statistical methods we combine the previous embedding vectors to form a synthetic feature matrix, then use the Gramian Angular Field and Markov Transition Field to augment the data information and extend the feature embedding by matrix integration. To verify the validity of our model in experiments, we implement multilayer perceptron as training model to realize prediction task. Our model has quite good interpretability so that it can well be understood what happens in the model framework and how the data is processed throughout the model pipeline. This can well help us make relative adjustments and tunes according to different problems and scenarios, thus get better prediction results.
暂无评论