Prediksi Tingkat Resiko Stroke Pasien Berbasis Gejala Klinik Menggunakan Multilayer Perceptron

Abstract

Stroke is a condition that represents one of the leading causes of death and permanent disability worldwide; therefore, early diagnosis is essential to minimize associated risks. This study aims to implement the Multilayer Perceptron (MLP) method on two different datasets to analyze the effect of the number of input attributes on stroke risk prediction performance. The study was conducted using two public datasets with variations in the number of records (35,000 and 70,000 instances) and the number of attributes (17 and 16 features). The experimental method was divided into two scenarios: Scenario 1 utilized all available attributes, while Scenario 2 employed only the intersecting attributes (two features) to evaluate model efficiency. Model performance was assessed using a confusion matrix with accuracy, precision, recall, and F1-score as evaluation metrics. The results indicate that the MLP model achieved accuracy ranging from 80% to 98% across both datasets. A significant finding shows that feature dimensionality reduction in Scenario 2, using only two input attributes, was still able to achieve accuracy above 80% on Dataset 2 and Dataset 3, demonstrating computational efficiency without a substantial decline in performance. This study demonstrates that MLP remains effective and efficient even with limited features, thereby highlighting that appropriate feature selection can improve model efficiency without significantly compromising performance. These findings are expected to serve as a reference for the development of artificial intelligence-based disease prediction systems in the future.