1. IoT-Enabled Smart Mental Health Assessment Using Deep Hybrid Regression Models Over Actigraph-Based Sequential Motor Activity Data.
- Author
-
Arora, Anshika
- Subjects
- *
REGRESSION analysis , *SLEEP quality , *MENTAL health , *STATISTICAL significance , *FEATURE extraction - Abstract
This research puts forward novel deep hybrid frameworks for assessment of mental health indicators viz. depression and sleep quality using IoT-based motor activity recordings. The study employs long short-term memory (LSTM) to extract high-level features from sequential motor activity data, which are then combined with statistical features of the raw data to form a hybrid feature extraction model. The combined feature vector obtained via the hybrid feature extractor is fed into four regression models, namely linear regression (LR), sequential minimal optimization regression (SMOR), random forest regression (RFR) and adaptive neuro-fuzzy inference system (ANFIS) forming four deep hybrid regression models, namely LSTM-LR, LSTM-SMOR, LSTM-RFR, and LSTM-ANFIS for prediction of depression and sleep quality. The proposed deep hybrid frameworks are validated on benchmark datasets, namely the Depresjon dataset and the MESA actigraphy dataset, and the best performance is observed by LSTM-RFR with the adjusted R2 value of 74.19% on the MESA dataset. On validating the significance of statistical features, it has been observed that all the models show significant improvement in performance on combining the statistical features with high-level features with highest reduction in mean absolute error as 4.0436 and highest increase in R2 statistic as 0.1651. Model building time of the four regression methods is also compared for which LR gave the best results as only 0.01 s for training and testing. [ABSTRACT FROM AUTHOR]
- Published
- 2024
- Full Text
- View/download PDF