Evaluasi Kinerja Algoritma Klasifikasi dalam Memprediksi Gangguan Paru pada dataset PEEP Expiratory Occlusion

Abstract

Asthma is a chronic inflammatory disease of the airways that poses a significant global health concern,
with a notable prevalence in Indonesia, affecting approximately 2.6% of the total population. The high risk
and the often subjective nature of conventional diagnosis necessitate the development of objective and
accurate early detection systems. This research focuses on analyzing and comparing the performance of
three prominent Machine Learning algorithms Random Forest (RF), Support Vector Machine (SVM), and
K-Nearest Neighbor (KNN) in classifying lung conditions, specifically Asthma versus Non-Asthma, utilizing
features extracted from patient respiratory signals. The dataset employed is the Respiratory Dataset from
PEEP Study with Expiratory Occlusion, comprising 78 subjects equally divided into 39 asthma patients
and 39 normal subjects. The data underwent feature extraction, incorporating crucial respiratory
parameters such as pressure, flow, and tidal volume, before being split into training (80%) and testing
(20%) sets. Model performance was comprehensively evaluated using Accuracy, Precision, Recall
(Sensitivity), and F1-Score metrics to ensure diagnostic reliability. Testing results demonstrate that the
Random Forest algorithm delivered the most outstanding and stable performance across all metrics,
achieving an Accuracy of 99,39%, F1-Score of 98,97%, Recall of 98,82%, and Precision of 99,13%. This
was followed by KNN with an Accuracy of 90,97%. In contrast, SVM with the default RBF kernel showed
the lowest performance, yielding an Accuracy of only 70,82% and a very low F1-Score of 2,56%, indicating
a significant failure in identifying most positive asthma cases (low Recall). In conclusion, the Random
Forest ensemble learning approach proves to be the most effective and reliable method for handling the
complex and variable characteristics of respiratory physiological data, making it the recommended model
for developing a robust early diagnostic system for asthma.