Teacher’s language performance prediction using multiresolution Sinusoidal Neural Networks optimized by Leaf-in-Wind algorithm

Accurately forecasting teacher language performance is critical to enhancing instructional quality and facilitating directed professional development. Current models have more limitations such as low generalization ability and inability to capture multiscale temporal and linguistic patterns. To overcome these challenges, this paper proposes a novel prediction framework that integrates Multiresolution Sinusoidal Neural Networks with Leaf-in-Wind Optimization algorithm (PTL-MSNN-LWO).The Multiresolution Sinusoidal Neural Networks (MSNN) effectively decomposes teacher performance data into sinusoidal components across multiple resolutions and predict their performance. To improve the model accuracy and robustness, the Leaf-in-Wind Optimization (LWO) algorithm used by dynamically optimizing MSNN parameters through a search mechanism that mimics the adaptive motion of a leaf swayed by wind forces. The proposed model was evaluated using Ed X Courses Dataset. The study also involved an augmented reality (AR) learning activity with nonprofessional students divided into experimental and control groups. While both used the same AR environment, only the experimental group received MSNN based deep learning support. The experimental results show that the PTL-MSNN-LWO achieved a prediction accuracy of 99.48 %, precision of 98.72 %, F1-score of 98.53 %, outperforming exiting methods. These results confirm the efficacy of the proposed method in efficient and reliable predictions for teacher language performance assessment.