AI-Driven Prediction of Lung Cancer Progression for Clinical and Strategic Healthcare Decision-Making

Abstract

Lung cancer is one of the most widespread and fatal types of cancer all over the world, and the early and precise diagnosis is essential to treat it successfully. The traditional approaches to diagnosing patients have been known to be unable to reveal the faint trends on patient data to enable effective decision making. New advances in artificial intelligence (AI) and machine learning (ML) allow finding predictive relationships and large-scale analysis of clinical data. In this paper, the ORANGE data mining platform was used to analyze a Kaggle lung cancer dataset where the characteristics of patients are age, gender, smoking status, and symptoms. To investigate the relationships between clinical phenomena and clinical outcomes, the predictive target chosen was survival. The best performing algorithm of the tested ones was the Decision Tree (CA = 0.931, AUC = 0.983, F1 = 0.928, Precision = 0.929, Recall = 0.931, MCC = 0.790). Random Forest classifier took the second place (CA = 0.912, AUC = 0.988, F1 = 0.903, Precision = 0.919, Recall = 0.912, MCC = 0.731), and kNN got moderate results. In comparison, Gradient Boosting (CA = 0.781) and Naiviste Bayes (CA = 0.780) also yielded much weaker results especially in terms of F1-score and MCC. The results have shown that Decision Tree and Random Forest have the most accurate classification behavior in predicting survival, and the rest of the algorithms report poorer discriminatory performance. On the whole, the findings indicate that predictive modeling powered by AI can be used to assist or refute the use of possible predictors to guide healthcare professionals and policymakers to determine the most significant variables in the diagnosis, treatment planning, and strategic plan in healthcare.