Real-time monitoring and optimization of traditional sports training load driven by IoT edge computing

To improve athletic performance and reduce risk of overtraining or injury, managing and optimizing training load will be critical to achieving this goal. Both traditional methods, which involved manual tracking and feedback, and existing cloud solutions that introduce latency and inadequate adaptive mechanisms to manage the noise of multimodal IoT sensor-generated data, limit the ability to make informed, rapid decisions. The purpose of this research is to develop a framework that predicts and optimizes traditional training load in real-time at the edge by using advanced deep-learning techniques in combination with IoT sensor data. The development of a Self-adaptive Long Short-Term Memory (SA-LSTM) network enhances dynamic temporal modeling and hyperparameter tuning with the addition of the Fossa Optimization Algorithm (FOA). This new framework collects multimodal Internet of Things (IoT) sports data including heart rate, acceleration, speed and movement. Preprocessing involves Savitzky-Golay filter, which removes noise, and Min-Max normalization, which scales features to the same range. Power Spectral Density (PSD) in the frequency domain is used to capture periodicity and energy density, both of which are required for assessing training loads. The attention mechanism enables the algorithm to prioritize critical time points for immediate prediction. Implementation is completed using Python, with TensorFlow/Keras, NumPy and Pandas, optimised for deployment to the edge. The proposed framework obtained an accuracy of 98.7% for data processing, an F1 score of 95.2%, precision of 90.1%, with 28 J of low energy consumption. Using the FOA, SA-LSTM is adapted better to sequential variability and hyperpapmeter tuning is accurate; thus, allowing the training load to optimally manage real-time observations and adjustments. The SA-LSTM-FOA framework supports real-time optimisation of training loads assisted by IoT providing low latency, high accuracy and energy efficiency for edge deployment.