Prediction and Visualization Analysis of Lung Cancer Risk by Machine Learning

Abstract

One of the most lethal cancer types in the world is lung cancer. It gravely threats human health and causes more than one million deaths annually. Current research demonstrate that an early and accurate determination of lung cancer is the key element in reducing the mortality rate, which motivates us to predict cancer at an early stage by exploiting the capacity of machine learning. In this work, both qualitative visualization analysis and quantitative experiments are conducted to reveal the relevant factors and the effectiveness of lung cancer risk prediction. Visual analyses are implemented to display the distribution and correlation between characteristics of patients and lung cancer risk. The visualizations demonstrate that smoking has a relatively large impact on lung cancer. Moreover, various types of machine learning algorithms are implemented and compared, including the logistic regression, K-nearest neighbors, random forest, and decision tree. These methods achieve satisfactory prediction results, with an accuracy of about 84% to 90%.