SafeMile
This work presents SafeMile, an AI-powered application designed to prevent running-related injuries. Running is a popular form of exercise, yet injuries are common, with most runners experiencing at least one injury during their training career due to factors such as overuse, insufficient recovery, or lack of strength. Using AI and machine learning, specifically support vector machines (SVM), SafeMile analyzes biomechanical and environmental factors to identify risk patterns. The app provides real-time feedback and personalized recommendations to improve running techniques and reduce injury risk.
SafeMile's goal is to create a personalized training program offering real-time injury prevention recommendations.
SafeMile System Flow.
The SafeMile system consists of two main components: a running watch for real-time feedback and data collection, and a mobile app for in-depth analysis and data input on a larger screen.
The system uses an SVM (Support Vector Machine) model to analyze patterns and assess injury risks. This enables SafeMile to provide real-time recommendations to reduce injury risk and enhance training quality.
All training and health data are stored in the cloud, allowing for long-term analysis and personalized training plans tailored to the user's progress and needs.
Through reinforcement learning, SafeMile dynamically adapts based on user feedback, such as reported pain after sessions. This helps the system adjust future recommendations to better prevent injuries and optimize training.
SafeMile AI model training.
To function effectively, the SafeMile AI model requires initial training. The data is divided into two partitions: training data for learning and test data for validation, ensuring the model can generalize to unseen data.
The model is trained on large amounts of historical running data, where each runner is classified by injury risk (injured = 1, not injured = 0) and various features such as biomechanics, running technique, and injury history. The model uses an SVM algorithm to predict injury risk by separating runners into high-risk and low-risk groups based on their individual data.
Model validation tests its ability to generalize to unseen data, ensuring it is not overfitted to the training data. This process confirms the model can provide reliable recommendations for new situations.
Once training and validation are complete, the model processes real-time data.
User Interface.
The user interface for the SafeMile system includes the running watch, mobile app, and real-time feedback (warnings) that I have developed.
