Bed Rise Prediction IoT System under Varying Blanket Conditions using Multimodal Machine Learning

In this paper, we propose bed rise prediction IoT system under varying blanket conditions using an accelerometer and a low-resolution infrared sensor as multimodal machine learning. The system predicts bed rise actions based on CNN-LSTM from the movements observed prior to a bed rise action, therefore we aim to reduce some risks, i.e., elderly individuals injuring themselves and falls. Furthermore, we investigate how combining two sensor data affects the results under three conditions: no blanket, thin blanket, and thick blanket. The result has demonstrated the accuracy improved slightly under the blanket conditions when using both, compared to using only infrared data. Additionally, we have indicated that the variability in prediction results based on the initial weight values can be reduced by only adjusting the learning rate.