No abstract

The lung cancer threat has become a critical issue for public health. Research has been devoted to its clinical study but only a few studies have addressed the issue from a holistic perspective that included social, economic, and environmental dimensions. Therefore, in this study, risk factors or features, such as air pollution, tobacco use, socioeconomic status, employment status, marital status, and environment, were comprehensively considered when constructing a predictive model. These risk factors were analyzed and selected using stepwise regression and the variance inflation factor to eliminate the possibility of multicollinearity. To build efficient and informative prediction models of lung cancer incidence rates, several machine learning algorithms with cross-validation were adopted, namely, linear regression, support vector regression, random forest, K-nearest neighbor, and cubist model tree. A case study in Taiwan showed that the cubist model tree with feature selection was the best model with an RMSE of 3.310 and an R-squared of 0.960. Through these predictive models, we also found that apart from smoking, the average NO2 concentration, employment percentage, and number of factories were also important factors that had significant impacts on the incidence of lung cancer. In addition, the random forest model without feature selection and with feature selection could support the interpretation of the most contributing variables. The predictive model proposed in the present study can help to precisely analyze and estimate lung cancer incidence rates so that effective preventative measures can be developed. Furthermore, the risk factors involved in the predictive model can help with the future analysis of lung cancer incidence rates from a holistic perspective.