An explainable artificial intelligence driven fall system for sensor data analysis enhanced by butterworth filtering

Abstract

The detection of falls is an essential component of healthcare monitoring systems, especially for older people at a greater risk of falling than younger people. To address the shortcomings of previously established methodologies, this research proposes a unique Artificial Intelligence driven sensor-based methodology that utilizes the SisFall dataset in conjunction with a Recurrent Neural Network - Long Short-Term Memory model. Two methods were considered: one using a Butterworth filter and the other without filtering. The results emphasize the significance of noise reduction in enhancing model performance. Additionally, the integration of Explainable Artificial Intelligence techniques brings transparency and interpretability to the model’s predictions, enhancing its dependability and trustworthiness in healthcare applications. Using Artificial Intelligence driven fall detection with Explainable Artificial Intelligence for transparent decision-making, this methodology presents a robust approach to improving accuracy and reducing false alarms in real-world healthcare settings. The study demonstrates that combining advanced filtering techniques with Explainable Artificial Intelligence algorithms successfully overcomes the challenges associated with traditional fall detection systems. The findings further confirm that the application of an Artificial Intelligence based Butterworth filter significantly enhances model accuracy, achieving 98.96% compared to 79.77% without filtering. These findings highlight the potential of Artificial Intelligence driven fall detection systems in healthcare, paving the way for more accurate, interpretable, and reliable monitoring solutions that can enhance elderly safety and improve real-time clinical decision-making.