Lung cancer prediction using machine learning

Lung cancer continues to be one of the main causes of cancer deaths all over the world, majorly because of the advanced diagnosis (Bertolaccini et al., 2017). The research project seeks to improve early detection by carrying predictive modelling with several different machine learning algorithms. Python was used to analyse the data set consisting of 309 entries of health-related survey questions age, gender, smoking pattern and signs and symptoms like cough, chest pain and shortness of breath. There are five models of classification which are Artificial Neural Network (ANN), Logistic Regression, Decision Tree, Naive Bayes and Gradient Boosting and compared each one. The performance of each model was measured using accuracy, precision, recall, F1 scores, confusion matrix and 10-fold cross-validation. The overall highest performance was achieved by Gradient Boosting (accuracy of 89.32%) and Logistic Regression obtained the most consistent cross-validation score (91.92%). The ANN had flawless recall but a major issue with the false positives. Decision Tree and Naive Bayes have proved also to be reliable, which makes their implementation into clinical measures to be possible. This study proves that machine learning algorithms can prove to be of major help in predicting lung cancer when tuned more to sensitivity or specifically (Raoof et al., 2020). Even though some of the models showed trade-offs (tendency of ANN to over-predict positive cases), combination of several algorithms and the highly directional parameter tuning can lead to better reliability of diagnostics. Predictive modelling is effective in early detection that not only raises the chances of survival but also minimizes the burden on the healthcare systems (Raoof et al., 2020).