Fuzzy Logic-Based Model for Stroke-Risk Prediction Using Non-Invasive Risk Factors: A Mamdani Inference Approach

Abstract

Stroke-risk assessment can support preventive intervention by identifying individuals whose combined demographic, behavioural and health-related characteristics may indicate elevated risk. This study develops a Mamdani fuzzy inference model for preliminary stroke-risk classification using eight non-invasive inputs: age, gender, family history of stroke, smoking history, alcohol consumption, body mass index (BMI), exercise frequency and hypertension status. Each input was represented by a crisp value mapped to linguistic categories, with triangular membership functions used to model the defined intervals. Age, gender, BMI and exercise frequency were assigned two linguistic labels each, while family history, smoking, alcohol consumption and hypertension were assigned three labels each, giving 34 × 24 = 1,296 possible antecedent combinations. A 1,296-rule IF–THEN base was formulated with input from a domain expert. The system was implemented in MATLAB using the Fuzzy Logic Toolbox and tested with one representative input vector, [0, 1, 1, 0, 0, 1, 1, 0]. The defuzzified output was 0.16, which falls within the defined Low Risk interval. The result demonstrates the feasibility of representing accessible risk factors with an explicit fuzzy rule structure. However, the study is a proof-of-concept: a single simulated case cannot establish predictive accuracy, discrimination, calibration, clinical utility or generalisability.  Independent rule-base review and validation against appropriately characterised patient data are therefore required before clinical application.