AI-Based Prediction of Systolic Blood Pressure Using Biometric and Clinical Data

Abstract

Early and accurate prediction of systolic blood pressure (SBP) is essential for preventing cardiovascular complications and improving patient care. In this study, we examined whether a Multi-Layer Perceptron (MLP) model could effectively estimate SBP by combining several biometric and clinical factors. The dataset involved 128 adults visiting a cardiology clinic and included variables such as age, abdominal circumference, glucose, lipid profiles, creatinine, urea, hemoglobin, hematocrit, and diastolic blood pressure (DBP). Before training, the data were carefully cleaned and normalized to ensure consistency. An MLP model with a single hidden layer was developed and evaluated using two data-split scenarios (70/30 and 80/20 for training and testing). Several activation functions were explored—sigmoid, hyperbolic tangent, and identity—to determine the most efficient setup. Interestingly enough, the model using sigmoid functions in both layers delivered the lowest testing error (MSE = 0.004) in the 80/20 split, suggesting strong predictive performance. The analysis revealed that DBP, abdominal circumference, and hemoglobin (HGB) played the most critical roles in prediction accuracy. In addition, urea, hematocrit (HCT), and creatinine showed consistent importance across models with testing errors below 0.046. Taken together, these results indicate that MLP-based models can be valuable, practical, and interpretable tools for SBP prediction. Incorporating such approaches into clinical practice could support personalized cardiovascular risk assessment and more informed decision-making for patient care.