Evaluasi Model Machine learning untuk Prediksi Keparahan Kanker Berdasarkan Data Real-world Global

Abstract

Cancer is one of the leading causes of death worldwide and places a significant burden on healthcare systems. Information on cancer severity is crucial for prioritizing treatment and resource planning. This study aims to develop and compare machine learning-based cancer severity classification models using global cancer patient data from 2015–2024. The dataset comprises 50,000 patients with various demographic, lifestyle, environmental, and clinical attributes, as well as severity scores (Target Severity Score). The dataset used in this study was obtained from the open data platform Kaggle (www.kaggle.com), which contains global cancer patient data from 2015 to 2024. The severity score is converted into a binary variable with two classes: low and high severity. The research steps include data preprocessing (cleaning, categorical transformation of variables with one-hot encoding, standardization), data division into training and testing data with a stratified 80:20 ratio, and the development of three classification models: Logistic Regression, K-Nearest Neighbors (K-NN), and Support Vector Machine (SVM) with RBF kernel. Model performance was evaluated using accuracy, precision, recall, F1-score, and confusion matrix, and validated with 5-fold cross-validation. Experimental results showed that Logistic regression achieved 99.82% accuracy, 99.86% precision, 99.78% recall, and 99.82% F1-score, with very small classification errors. SVM achieved 98.22% accuracy with also high performance, while K-NN only achieved an accuracy of around 79.42%. Cross-validation results confirmed that Logistic regression had the highest average accuracy and the most stability. Thus, Logistic regression is recommended as the primary model for predicting cancer severity in this dataset and has the potential for further development as a component of a clinical decision support system