Quest for Ground-Truth or Stochastic Myth by Leveraging the AI-Powered Wearable Device for Dementia Disease Detection: A Pilot Study

Abstract

Dementia affects over 50 million people worldwide, with numbers expected to
triple by 2050. Traditional diagnostic methods often lack early detection
capabilities and real-time monitoring, leading to delayed interventions and
increased caregiver burden. This study aimed to develop an integrated dementia
detection system combining wearable Internet of Things (IoT) with deep learning
for early identification and continuous monitor of dementia. The system has three
core components: (1) a wearable IoT using ESP32 and MAX30102 sensors to
collect data, (2) a tree-based, stacked learning approaches with 3-base classifiers
(decision tree, random forest and adaboost) – and a XGBoost meta-classifier, and
(3) a mobile app to ease data visualization. The dataset comprised of 2,149 records
with 16-features. Preprocessing handled missing values to ensure data
quality/integrity – while, normalization was used to address imbalanced dataset.
Results showed that the stacked model yielded a 99.7% accuracy, 100% sensitivity,
99.4% specificity, and 99.8% AUC. While IoMT device successfully collected
physiological data as displayed over mobile app – model shows that the AIPowered
unit can effectively help detect dementia.