Machine Learning-Based Fall Detection Algorithm Using Data Fusion Techniques

For} elderly people, accurate fall detection offers prompt medical assistance. This study describes a reliable fall detection system that combines multi-level data with machine learning. The SisFall dataset, which is open to the public, was chosen because of its benefits. Butterworth filtering and sliding window segmentation were used as preprocessing techniques. There were three degrees of data fusion used: data-level fusion (Kalman filtering) fused accelerometer and gyroscope signals; feature-level fusion (PCA) aggregated extracted features; and decision-level fusion (weighted majority vote bagging) combined classifier outputs. XGBoost, Naive Bayes, Random Forest, SVM, KNN, and logistic regression were among the machine learning models that were assessed. Accuracy and robustness were improved through the use of data fusion techniques and machine learning algorithms. The effectiveness of fusing complementary data at various levels was demonstrated by the final multi-level fusion technique, which had a fall detection accuracy of 99.0%.