Research On Student Achievement Prediction Model Based On Ensemble Learning Algorithm

Student achievement prediction is an important research object in educational data mining.Student achievement is also an important factor in the evaluation of students’ comprehensive quality.Therefore,by predicting students’ grades,students with learning problems can be screened out early.It is very important for teachers to conduct daily teaching management work and students to obtain good academic results to give corresponding guidance to these students in a timely manner.Due to the differences in teaching methods,student habits,and software and hardware levels among schools in different regions,the data collected by different schools have obvious differences in data formats,data types,and data richness.Differences in schools’ collection of student academic data have led to large differences in the data used in related studies.Current research on student achievement prediction pays less attention to the correlation between course knowledge point data and grades,and its applicability and expansibility are narrow,which is not conducive to the practical application of relevant research results.At the same time,the existing research also pays less attention to the impact of time series factors on the prediction of student achievement.In response to the above problems,this thesis proposes two student achievement prediction models based on course knowledge points and student historical achievement data sets.The specific research contents are as follows:(1)The characteristic information of students’ examination is constructed by using the knowledge point information of the course,the examination point information of the examination paper and the history examination record of the students.Construct the characteristic fields that can show the changing trend of students’ examination scores,so that the model can capture the dynamic changes of students’ examination scores and enhance the accuracy of the model in time series prediction.(2)Based on attention mechanism and long-term and short-term memory network,an Att-LSTM achievement prediction model is proposed.The introduction of attention mechanism can enhance the ability of long-term and short-term memory network models to select key information from a large number of input feature data,focusing on the characteristic information that has a great impact on students’ achievement.The experimental results show that,compared with the single long-and short-term memory network achievement prediction model,the Att-LSTM model can reduce the error in the RMSE index.At the same time,this thesis also analyzes the influence of the trend characteristic field of students’ examination scores on the prediction results of the AttLSTM model,and the results show that the trend characteristic field can improve the prediction effect of the model.(3)A student achievement prediction model based on multi-model superposition based on integrated learning Stacking method is proposed.The Stacking model consists of two layers of learning models.The first layer is composed of integrated learning frameworks Light GBM,XGBoost and Att-LSTM model,and the second layer is composed of logical regression algorithms.Through the experiment on the data set,the experimental results show that the student achievement prediction model based on integrated learning Stacking is better than the single achievement prediction model in terms of average absolute error,root mean square error and determination coefficient.The research in this thesis shows that the research on student achievement prediction based on the curriculum knowledge point dataset has good accuracy.At the same time,the multi-model overlay student achievement prediction model based on ensemble learning Stacking can reduce the risk of model overfitting and enhance the generalization performance of the prediction model.