This paper proposes a student performance prediction model based on multidimensional time-series data analysis (MTAPSP) to predict student performance in online learning. The model integrates students' learning behaviors, assessment scores, and demographic information to capture the relationship between multiple factors and student performance. The model uses a multi-layer LSTM network with a multi-head self-attention (MHSA) mechanism to extract time-series features and enhance the model's ability to predict student performance in multi-classification. Additionally, an artificial neural network (ANN) is used to integrate temporal behavioral features, assessment score features, and demographic data to improve the model's performance. The model was tested on the Open University Learning Analytics Dataset (OULAD) and achieved 74% accuracy and 73% F1 scores in a four-category prediction task, and 99.08% accuracy and 99.08% F1 scores in an early risk prediction task. The results show that the model outperforms the benchmark model in both multi-classification prediction and early prediction ability. The model's early prediction capability allows instructors to identify at-risk students early and intervene in a timely manner, improving student performance and learning outcomes. The study also highlights the importance of considering demographic information and the impact of assessment scores on student performance. The model's ability to predict student performance at different stages of the course provides valuable insights for educators to develop targeted instructional programs and support students effectively.This paper proposes a student performance prediction model based on multidimensional time-series data analysis (MTAPSP) to predict student performance in online learning. The model integrates students' learning behaviors, assessment scores, and demographic information to capture the relationship between multiple factors and student performance. The model uses a multi-layer LSTM network with a multi-head self-attention (MHSA) mechanism to extract time-series features and enhance the model's ability to predict student performance in multi-classification. Additionally, an artificial neural network (ANN) is used to integrate temporal behavioral features, assessment score features, and demographic data to improve the model's performance. The model was tested on the Open University Learning Analytics Dataset (OULAD) and achieved 74% accuracy and 73% F1 scores in a four-category prediction task, and 99.08% accuracy and 99.08% F1 scores in an early risk prediction task. The results show that the model outperforms the benchmark model in both multi-classification prediction and early prediction ability. The model's early prediction capability allows instructors to identify at-risk students early and intervene in a timely manner, improving student performance and learning outcomes. The study also highlights the importance of considering demographic information and the impact of assessment scores on student performance. The model's ability to predict student performance at different stages of the course provides valuable insights for educators to develop targeted instructional programs and support students effectively.