Original Research · Open access
Dementia Detection Unit Via A Wearable Internet Of Things Fused With Deep Machine Learning Scheme
Abstract
Dementia affects over 50 million people globally, with numbers expected to triple by 2050. Traditional diagnostic methods lack early detection capabilities and realtime monitor, leading to delayed intervention and increased caregiver burden. Our study deploys an integrated dementia detection unit combined with wearable Internet of Things (IoT) and deep machine learning model for early identification and continuous monitor of dementia symptoms. A comprehensive system was designed incorporating three components namely: (a) a wearable IoT device using MAX30102 sensor and the ESP32 microcontroller for physiological data collection, (2) a hybrid CNN-LSTM deep learning ensemble, and (c) a mobile app interface for data visualization. The dataset comprised 1,510 patient records with 11 features including demographic, cognitive, and clinical variables. Data preprocessing involved missing value handling, categorical encoding, feature scaling, and SMOTE balancing to address class imbalance. The Multi-Layer Perceptron showed superior performance on the balanced dataset with 0.97 Accuracy, 1.00 Sensitivity, 0.94 Specificity, and 0.98 AUC-ROC. The wearable device successfully collected real-time physiological data including blood glucose, blood pressure, and cholesterol levels – with our app providing seamless integration between hardware and software components, enabling realtime monitor alert and prediction delivery. The integrated system yields a significant advancement in dementia care with early detection abilities, continuous monitor, and improved accessibility. The MLP model's exceptional performance, combined with the practical wearable implementation, provides a scalable solution for healthcare systems seeking to improve dementia diagnosis and management while reducing caregiver burden.
Keywords
Cite this paper
Choose a citation style, then copy the formatted reference for your manuscript or academic record.
Onoma, P., Ako, R. E., Okpako, E. A., Oghorodi, D., Abere, R. A., Okpor, M. D., Yoro, R. E., Geteloma, V. O., Ibor, A. E., Max-egba, A. T., & Star, N. U. (2025). Dementia Detection Unit Via A Wearable Internet Of Things Fused With Deep Machine Learning Scheme. International Journal of Advanced Computing and Intelligent System, 4(1), online first. https://doi.org/10.66917/ijacis.a000022
Onoma, Paul, et al. “Dementia Detection Unit Via A Wearable Internet Of Things Fused With Deep Machine Learning Scheme.” International Journal of Advanced Computing and Intelligent System, vol. 4, no. 1, 2025, pp. online first. doi:10.66917/ijacis.a000022
Onoma, Paul, Rita Erhovwo Ako, Ejaita Abugor Okpako, Duke Oghorodi, Reuben Akporube Abere, Margaret Dumebi Okpor, Rume Elizabeth Yoro, Victor Ochuko Geteloma, Ayei Egu Ibor, Asuobite Thankgod Max-egba, and Niemogha Umiey Star. “Dementia Detection Unit Via A Wearable Internet Of Things Fused With Deep Machine Learning Scheme.” International Journal of Advanced Computing and Intelligent System 4(1) (2025): online first. https://doi.org/10.66917/ijacis.a000022
Onoma, P.; Ako, R. E.; Okpako, E. A.; Oghorodi, D.; Abere, R. A.; Okpor, M. D.; Yoro, R. E.; Geteloma, V. O.; Ibor, A. E.; Max-egba, A. T.; Star, N. U. (2025) ‘Dementia Detection Unit Via A Wearable Internet Of Things Fused With Deep Machine Learning Scheme’, International Journal of Advanced Computing and Intelligent System, 4(1), pp. online first. doi:10.66917/ijacis.a000022
BibTeX
@article{msisijacis-a000022,
title = {Dementia Detection Unit Via A Wearable Internet Of Things Fused With Deep Machine Learning Scheme},
author = {Paul Onoma and Rita Erhovwo Ako and Ejaita Abugor Okpako and Duke Oghorodi and Reuben Akporube Abere and Margaret Dumebi Okpor and Rume Elizabeth Yoro and Victor Ochuko Geteloma and Ayei Egu Ibor and Asuobite Thankgod Max-egba and Niemogha Umiey Star},
journal = {International Journal of Advanced Computing and Intelligent System},
year = {2025},
volume = {4},
number = {1},
doi = {10.66917/ijacis.a000022}
}Related articles
Browse this journal
IJACIS
Lung Cancer Detection Using A Hybridized Contrast-Based Xception Model On Image Data: A Pilot Study
Lung cancer has since become the second largest direct cause of death when compared to other forms of cancer. Lung cancer is easily diagnosed via medical x-rays, MRIs and CT-scans. The use of learning schemes in identification of disease has become a veritable tool in medical frontiers. But, a major challenge with image classification – is attributed to the disease severity, and model’s ability to address feature enhancements in model designs for image processing. The models must capture intrinsic features with image enhancements in regions of interest – amongst other feats. Our study uses a median filter (AMF) with contrast enhancement (CLAHE) to normalize images, use a variation-mode decomposition as preprocessor to sustain multimodal image learning upheld by the Xception model. VMD handles complex image data to resolve mode mixing in its AMF and CLAHE. This is further upheld via localized context and hierarchical image representation in our CNN-based Xception model to yield improved training efficiency, model stability and generalization even with complex spatiotemporal dependencies inherent the image dataset. Our result shows the proposed method performs well in lung cancer detection, with significantly enhanced time convergence and reliability to yield a 1.000 Accuracy, 0.994 Precision, 0.989 Recall and 0.989 F1-score. Thus, the proposed model provides an effective image recognition and detection solution for lung cancer.
IJACIS
Attrition Rate Prediction Using A Frequency-Recency-Monetization-Based Smoteen-Boosted Approach
Businesses today seek to generate more revenue for their products and services. However, attrition or churn for products and services over the digital market space has continued to plague businesses to proffer consumers with various alternates and choices to services and products for competitive money’s worth. Today’s businesses must thus, seek to yield better value while meeting consumer needs cum demands through a retention strategy. Thus, our study utilizes the recency-frequency-monetization as retention model that explores SMOTEEN-based AdaBoost ensemble to study the impact of imbalanced dataset to identify consumer churn. With chi-square feature selection, our proposed model yields F1 0.9819, Accuracy 0.9882, Recall 0.9848, Precision 0.9886, and Specificity 0.9824 respectively with 9632-correctly identified and 15-incorrectly classified cases. Conversely, other scheme yields ranges of F1 [0.6305–0.8875], Accuracy [0.6432-0.8879], Precision [0.6320–0.8805], Recall [0.6128-0.8876] and Specificity [06370-0.8894] respectively. This shows that model yields enhanced performance, and that the explored RFM-retention strategy was found to curb churn and improve monetization.
IJACIS
A Smart One-Stop Business Solution Using Flutter
In this paper, a unique one-stop smart application that integrates note-taking, selling, and purchasing features into a single platform is introduced. Developed with Flutter, MongoDB, NodeJS, and Express, the application provides an easy-to-use interface for smooth participation in e-commerce operations. Its unique feature is that it combines the features of a buyer, seller, and guest mode, making it easier for users to conduct transactions and effectively manage their accounts. The note-taking feature of the app, which is available in all other modes, makes it much easier for users to take notes while managing their purchases or shopping. The software protects user data by using SQL-Lite for local data storage, which lowers the possibility of unwanted access or data breaches. The app's architecture also permits scalability and future growth, with the intention of introducing a delivery portion that will generate revenue for authorized staff members. This development supports the app's objective of providing a complete solution for buyers and sellers, doing away with the need for numerous platforms for various tasks. Compared to cloud-based solutions, the app's usage of local data storage reduces its carbon footprint in terms of environmental impact. This emphasis on environmental sustainability is consistent with current developments in technology, where eco-friendly methods are becoming more and more important. Overall, this initiative offers a forward-thinking method of approaching digital commerce by providing a safe, adaptable platform that takes into account the sustainability of the environment and the changing demands of users.
IJACIS
Comparative Study On Australia Weather Classification
Classification is a key data analysis technique for developing predictive models from labeled data. This study compares the performance of four classification algorithms such as Decision Tree, Neural Network, Naïve Bayes, and Support Vector Machine, applied to the Australian weather dataset. The objective is to predict the next day’s rainfall based on the target variable RainTomorrow. Each model is trained and evaluated using a consistent preprocessing pipeline and evaluated with various performance metrics including accuracy, precision, recall, and F1-score. Our analysis reveals Support Vector Machine (SVM) as the top-performing classifier, offering practical insights for weather-based predictive modeling tasks.