A Comparative Analysis of Regression Algorithms in Lung Cancer Prediction

Results of All Cancers The mortality of lung cancer is highest among both international and domestic patients. Hence, early detection and accurate prediction are essential for treatment-affecting diagnoses to improve patient outcomes. Throughout the evolution of machine learning technologies, numerous regression algorithms have been introduced to predict lung cancer using clinical data or imaging datasets. In this work, we perform a comparison analysis among linear regression, logistic regression, and all commonly used decision-tree-based methods, including Decision Trees (CART), Random Forests, and Support Vector Machines for predicting lung cancer diagnosis and prognosis. We use patient data from two hospitals, including age, smoking information, lung function tests, and imaging results, to perform the analysis. Our results suggest that random forest and support vector machines generally offer the best accuracy, sensitivity, and specificity in diagnosis and prognosis prediction. Decision tree algorithms also do but depend heavily on the feature set chosen. In addition, we show improved performance in predicting placebo response by combining different data sources. Our study demonstrates that selecting regression algorithms are critical in lung cancer prediction and that combining multiple data types can improve accuracy. Realizing this potential requires accurate prediction algorithms, which are not currently available; such information can lead to better-developed forecast models and improved lung cancer patients' diagnoses and outcomes.