Hybrid spatio-temporal transformer and variational autoencoder framework for advanced sports injury prediction and analysis

Sports injuries significantly affect sports performance and overall health, making using accurate and reliable prognostic models imperative. Conventional algorithms for injury prediction, e.g., random forest and XGBoost, generally cannot understand intricate spatial and temporal patterns in the data and, therefore, provide low accuracy and suffer from poor generalizability. To defeat such limitations, this work introduces a new spatio-temporal transformer variational autoencoder (STT-VAE) structure, which combines state-of-the-art spatio-temporal modelling with the learning of the representation in the latent space. The integrated framework captures spatial and temporal relationships and tackles data variability remarkably with variational autoencoder-based regularization. The model is evaluated using the publicly available athlete injury and performance dataset, consisting of 200 athlete records with 17 attributes. These include demographic variables (age, gender, height, and weight), workload indicators (training hours per week, match count and training intensity), recovery-related variables (recovery days and rest intervals), performance scores, and a binary injury label. These features represent real-world athlete monitoring data and define the experimental context and practical scope of the proposed framework. It aligns with the theme of cybersecurity challenges in IoT-driven geospatial information systems, focusing on emerging security risks and protection strategies for IoT-enabled geospatial data environments. The novel STT-VAE architecture achieves a remarkable classification accuracy of 99.56%, which is much higher than baseline methods such as random forest (95.5% accuracy) and XGBoost (97% accuracy), with differences of 4.06% and 2.56%, respectively with precision (99.45%), recall (99.46%), F1-score (99.3%), and AUC-ROC all further attest to the accuracy and dependability of the novel architecture.