Supervised learning for prediction of personal factors associated with hemoptysis in patients with lung carcinoma

Background: Hemoptysis, as a symptom of lung cancer, could increase in intensity due to personal and lifestyle factors. Objective: Determine the ability of supervised learning to predict and classify personal factors predictive of hemoptysis in patients with lung carcinoma. Method: Analytical and cross-sectional study, from a secondary database of 1000 patients with lung cancer from data.world. The variables were age, obesity levels, dust allergy, consumption of alcoholic beverages, exposure as passive smoker and cigarette consumption. Student’s t test, Spearman correlation, decision tree and multilayer perceptron were used. Results: Men with lung cancer had higher rates of hemoptysis and personal exposure than women. Hemoptysis was moderately and positively correlated with alcohol consumption, dust allergy, obesity, and passive smoking. The decision tree correctly classified 83.50% as mild hemoptysis, 88.90% as moderate, and 97.30% as severe. The multilayer perceptron correctly predicted 92% of cases of mild hemoptysis, 97.70% of moderate and 100% of severe. Conclusions: Supervised learning models are accurate to correctly classify and predict personal and lifestyle factors associated with hemoptysis due to lung carcinoma.