Resolving Data Overload and Latency Issues in Multivariate Time-Series IoMT Data for Mental Health Monitoring

Pervasive healthcare services have evolved substantially in the recent years with IoMT rapidly changing the pace and scale of healthcare delivery. A promising application of IoMT is to fetch patterns of mental behaviour symptomatology based on bio-signals and transfer it to the corresponding hospital or psychologist for remote monitoring. But the data volume & performance, device diversity & interoperability, hacking & unauthorized use and acceptance & adoption barriers still restrain the practical and competent use of these devices. This research presents a plausible solution to surmount the data overload and processing latency in real-time sensory data collected through wearable devices for mental health monitoring. We propose a modified k-medoid data clustering technique based on time-frame restricted intra-cluster similarity calculations to obtain a summarized version of the original benchmark WESAD dataset for which the degree of information lost is minimum. A CNN is then trained on this summarized dataset for classification of mental state into the baseline, stress and amusement categories. The results show a significant reduction in the average execution time by 34% with a comparable accuracy to the original dataset, thus offering prompt real-time healthcare analytics.