4.Model-driven and data-driven disease classification

Utilizing a multifaceted approach, we employ cutting-edge AI and ML techniques, including Convolutional Neural Networks (CNN), Long Short-Term Memory (LSTM) networks, Random Forest (RF), Decision Trees (DT), and ensemble methods, to analyze longitudinal, time-series data derived from wearable accelerometers to classify various diseases based on patterns extracted from the activity data.

In addition to AI/ML models, statistical techniques such as cosinor analysis and Singular Spectrum Analysis (SSA) are integrated into the analysis pipeline. Cosinor analysis allows for the extraction of circadian rest-activity patterns in the activity data. SSA, on the other hand, decomposes the time series into its underlying components, revealing hidden trends or anomalies indicative of specific health conditions or diseases.

This approach holds potential in healthcare, as it enables early detection and precise classification of diseases using non-invasive, continuous monitoring methods.