This research proposal investigates how machine learning can forecast injury risk in cricket fast bowlers, who face some of the highest injury rates in the sport. Traditional workload metrics like the acute:chronic workload ratio (ACWR) offer limited predictive power, as they overlook individual biomechanics and contextual factors. The related work highlights a progression from rule-based thresholds to ensemble models and time-series deep learning, showing that while ensemble methods like Random Forests improve accuracy, LSTMs capture how risk builds over time. Advances in wearable sensors were also identified as critical, offering biomechanical insights that, when combined with workload data, can create more personalised and reliable injury predictions.

The proposal outlines a system that integrates workload logs (e.g., deliveries, rest periods, session intensity) with biomechanical features from wearable devices such as IMUs, measuring factors like trunk rotation velocity and front-foot impact forces. These data streams would be processed into multivariate time series and engineered into physiologically meaningful features, then used to train ensemble and LSTM-based models. To handle the rarity of injury events, methods like focal loss, oversampling, and cross-player validation are proposed to ensure robust generalisation. The system aims to deliver timely, interpretable risk scores, giving coaches and physiotherapists actionable early warnings and shifting injury prevention from reactive treatment to proactive management.