Performance evaluation and occupational health safety analysis of university teachers based on time-series feature...

This study proposes a novel framework for evaluating university teachers' performance and occupational health safety using advanced time-series feature extraction techniques. Traditional methods often fail to capture the temporal complexity and dynamic interdependencies of performance and health-related indicators. The proposed framework addresses these challenges by integrating domain-specific modeling, deep learning architectures, and temporal fusion strategies.

The framework comprises three major components: preliminaries, the Temporal Occupational Health Safety Evaluation Model (TOHSEM), and the Dynamic Temporal Feature Integration Strategy (DTFIS). The preliminaries define the mathematical foundation for modeling multivariate temporal sequences such as workload, stress signals, teaching evaluations, and physical activity. These variables are structured as time-indexed inputs for downstream processing. TOHSEM utilizes recurrent neural networks (RNNs) enhanced with attention mechanisms to identify critical patterns and dynamic dependencies within the data. The attention layers enable the model to weigh specific time steps and features based on relevance, facilitating the detection of performance fluctuations or health risks. This allows for more nuanced interpretations of temporal behavior. DTFIS implements a multi-scale approach, integrating short-term variations and long-term trends through hierarchical temporal representation. Temporal pooling and weighted fusion methods preserve both local responsiveness and global consistency. Additional components include normalization techniques for non-stationary sequences, outlier detection modules, and real-time updating pipelines.

Empirical validation is conducted using diverse data sources, including institutional logs, physiological monitoring, and subjective assessments. These datasets demonstrate the model's capacity to capture complex dynamics and support applications such as scheduling optimization, early risk detection, and personalized resource allocation.