Predicting Falls using Time Series Data from the Equilivest Device

"Falls pose a significant health risk to older adults, often resulting in serious injuries and reduced independence. This study explored two machine learning approaches for detecting falls using data from wearable sensors: a supervised binary classification approach trained on labeled fall and non fall data and an unsupervised anomaly detection approach trained exclusively on normal gait patterns. The results show that both approaches can accurately detect fall events within the scope of the study. The supervised models—Random Forest, Support Vector Machine, and Logistic Regression—demonstrated consistent performance, whereas the unsupervised One-Class Support Vector Machine (OCSVM) effectively identified anomalies without relying on fall data. This study offers a practical foundation for building fall detection systems and highlights the potential for future developments in predictive and real-time monitoring solutions. The motivation behind this dual approach lies in its long-term significance: if robust models can be developed to reliably detect fall events, they will provide a foundation for future work on more complex systems capable of predicting falls before they occur. Therefore, establishing dependable detection is a critical step toward enabling proactive and preventive safety solutions."