FRAILTY CLASSIFICATION

Frailty is a growing health concern in aging populations, linked to increased risks of disability, hospitalization, and mortality. Early and accurate classification into nonfrail, prefrail, and frail stages is essential for timely intervention. This project introduces a robust machine learning and deep learning-based system for frailty detection using gait parameters captured from Inertial Measurement Unit (IMU) sensors. Using the Kaggle dataset of key spatiotemporal features such as gait speed, stride length, cadence, postural sway, and asymmetry index are extracted for classification.

To address data imbalance, the SMOTE technique is applied. Four models Random Forest, XGBoost, Support Vector Machine (SVM), and Multilayer Perceptron (MLP) are trained, achieving accuracies up to 99%. Feature selection using KBest and Random Forest importance further enhances performance.

The system is implemented as an interactive Flask-based web application that supports secure login, model selection, input validation, and real-time prediction, offering a scalable and accessible frailty assessment tool.

Keywords: Frailty Classification, IMU Sensors, Machine Learning, Deep Learning, Flask Application, Gait Analysis, SMOTE, Elderly Health, Feature Selection.