Scholarlink

<p>Student performance prediction is a central educational challenge, providing useful information for targeted academic support and intervention. Existing models do not work well in incorporating behavioral aspects like study habits and attendance into the models, despite the potential impact of these factors on academic outcomes. The present paper suggests a hybrid deep learning model for predicting college students’ learning outcomes based on incorporating academic, behavioral, and demographic information. Conventional student performance prediction models tend to concentrate on academic grades, tending to overlook behavioral aspects like study habits, attendance, and class participation. The suggested framework addresses this weakness by utilizing multiple data sources, offering a more comprehensive prediction of student performance. The architecture utilizes ResNet for hierarchical feature extraction, Autoencoders to reduce dimensions, and Long Short-Term Memory (LSTM) networks to select features, extracting both subtle patterns and temporal relationships in the data. Hyperparameters are optimized via the Cheetah Algorithm to improve model efficiency and reduce overfitting. Through the synthesis of these sophisticated methods, the architecture attains improved prediction precision, providing actionable insights into student success determinants. The suggested hybrid deep learning model obtained a high accuracy of 98.6%, precision of 98.2%, and recall of 97.4%, showing high potential to identify high and low-performing students. The model was also effective in separating levels of student performance, as seen through its AUC-ROC score of 0.99, and accounted for 96% of variance in student scores with an R2 of 0.96. Moreover, the framework had low error predictions with an RMSE of 2.1 and MAE of 1.8, which clearly indicates that the framework is an extremely effective instrument for predicting the academic performance of students. This method offers a strong framework to predict student performance, with wider implications for early intervention and tailor-made academic interventions. It also highlights the benefit of incorporating behavior data into models of prediction in education.</p>

错误：搜索内容不能为空，请输入英文关键词
错误：关键词超出字数限制，请精简
高级检索

Application of deep learning to the development of a prediction model for college students’ learning outcomes

  • 期刊论文

  • 出版模式：

    开放获取

  • 发表日期：

    2025年12月16日

查看全文
  • Rui Ma,
  • Liu Xiao

摘要

Student performance prediction is a central educational challenge, providing useful information for targeted academic support and intervention. Existing models do not work well in incorporating behavioral aspects like study habits and attendance into the models, despite the potential impact of these factors on academic outcomes. The present paper suggests a hybrid deep learning model for predicting college students’ learning outcomes based on incorporating academic, behavioral, and demographic information. Conventional student performance prediction models tend to concentrate on academic grades, tending to overlook behavioral aspects like study habits, attendance, and class participation. The suggested framework addresses this weakness by utilizing multiple data sources, offering a more comprehensive prediction of student performance. The architecture utilizes ResNet for hierarchical feature extraction, Autoencoders to reduce dimensions, and Long Short-Term Memory (LSTM) networks to select features, extracting both subtle patterns and temporal relationships in the data. Hyperparameters are optimized via the Cheetah Algorithm to improve model efficiency and reduce overfitting. Through the synthesis of these sophisticated methods, the architecture attains improved prediction precision, providing actionable insights into student success determinants. The suggested hybrid deep learning model obtained a high accuracy of 98.6%, precision of 98.2%, and recall of 97.4%, showing high potential to identify high and low-performing students. The model was also effective in separating levels of student performance, as seen through its AUC-ROC score of 0.99, and accounted for 96% of variance in student scores with an R2 of 0.96. Moreover, the framework had low error predictions with an RMSE of 2.1 and MAE of 1.8, which clearly indicates that the framework is an extremely effective instrument for predicting the academic performance of students. This method offers a strong framework to predict student performance, with wider implications for early intervention and tailor-made academic interventions. It also highlights the benefit of incorporating behavior data into models of prediction in education.