Hybrid Classification Model Development for Enhancement the Data Driven Based Human Falling Detection Performance

Abstract
It is thus important to identify the cause of a fall in order to be able to protect the lives of people, especially the elderly and hospitalized patients who are most likely to fall. Three fall detection models are compared in this paper: These are the auto-encoder (AEC), the feed forward neural network (FFNN), and the particle swarm optimization-feed forward neural network (PSO-FFNN). The idea is to assess how well these models identify falls, perform, and are suitable for deployment in real-life scenarios. This study was based on data from 17 candidates, and the four classes of falls that were considered here include: Based on this dataset, the models were trained and evaluated with the emphasis made on the correct classification of falls and their differentiation from other movements. From the results presented, all the three models show encouraging signs in fall detection. The PSO-FFNN model, with 98% of accurate optimal rate, however, gave the highest accuracy level. By this result, it is clear that the PSO-FFNN model is highly reliable and accurate in detecting falls. The objective of this research was to develop more precise and credible forecasts of various types of motion turbulence employing the autoencoder’s unsupervised learning and FFFNN’s partitioned classification approach.

Author
Sazan Kamal Sulaiman

DOI
https://link.springer.com/chapter/10.1007/978-981-96-2468-3_77

Publisher

ISSN
1876-1119 1876-1100

Publish Date: