Evaluation of human vital parameters in smart cities based on machine learning

نوع: Type: Thesis

مقطع: Segment: PHD

عنوان: Title: Evaluation of human vital parameters in smart cities based on machine learning

ارائه دهنده: Provider: Nejood Abdulsattar

اساتید راهنما: Supervisors: Prof. Dr. Hassan Khotanlou , Prof. Dr. Hatam Abdoli

اساتید مشاور: Advisory Professors:

اساتید ممتحن یا داور: Examining professors or referees: Dr. Reza Mohammadi, Dr. Mahlagha Afrasiabi, Dr. Davar Giveki

زمان و تاریخ ارائه: Time and date of presentation: 2026

مکان ارائه: Place of presentation: سمينار گروه کامپيوتر

چکیده: Abstract: Healthcare professionals face numerous challenges when analyzing data and providing treatment, such as the parameters to be measured, the frequency of measurement, the responsibility of new medical devices to monitor patient health, and the difficulty in interpretation. Machine learning (ML) techniques are more efficient predictive models used to improve early prediction of patient care, reduce the cost of implementing healthcare systems, and ease interpretation. This study proposes a new model to merge temporal prediction and interpretation by using multiple agent systems to enhance the confidence of medical staff in the Artificial Intelligence systems. The new model deals with eight agents. A data agent for input distribution, the LR, AE, and HMM agent for accurate temporal prediction, and the RF, SHAP agent to improve the accuracy and interpret the results. Then the coordinator agent documents all the phases that lead to the final decision and uses R2, risk probability rules, and SHAP interpretation for the final clinical status give important recommendations and sent notification to the supervisory agent in the error or risk states. Finally, the supervisory agent decides when the model must be retrained, corrects the model when it fails, manages the policy, and records all decisions. Three datasets are used: the first dataset, the MIM-IC-II database of the MIT PhysioBank archive, which contains 1023 patients’ records; the second dataset, the EHR dataset that contains 10000 patients’ records; and the third dataset, the cardiovascular disease dataset collected from Kaggle, refer to a dataset of patient samples focused on detecting cardiovascular problems. The proposed model is distinguished from traditional methods in its advanced system that combines the acts of several agents and the intelligent distribution of responsibilities among them, characterized by strong flexibility, reliability, dealing with big datasets, and short response time, giving interpretations, recommendations, and alarms in risks and predicting the clinical states with high accuracy. The new model achieved an accuracy of 98.4%, a precision 95.3%, a sensitivity of 99.2%, a specificity of 99.1%, an F1-Score of 97.1%, and an R2 of 98% when the MIMIC-II dataset is used. While achieving an accuracy of 93%, a precision of 92%, a recall of 94%, an F1-Score of 93%, an AUC-ROC of 94%, and an AUC-PR of 89% when the EHR dataset is used. While shown, age, id, and AP-LO are the most important attributes in predicting diastolic blood pressure. Where age had (-157), id (+17.53), and ap-lo (-213.03) effects. While, gender, height, and glu were the most important attributes in predicting systolic blood pressure. Where gender had (-4.41), height (+2.78), and glu (+2.48) effects when the cardiovascular disease dataset is used.