A comparative analysis of classical machine learning and deep learning techniques for predicting lung cancer survivability

• Main Authors: Huang S., Arpaci I., Al-Emran M., K?l?�arslan S., Al-Sharafi M.A.
• Other Authors: 56465178700
• Format: Article
• Published: Springer 2024
• Subjects: Classical machine learning; Deep learning; Demographic and clinical features; Lung cancer; Survival period prediction; Biological organs; Computerized tomography; Decision trees; Deep neural networks; Diseases; Learning algorithms; Learning systems; Population statistics; Supervised learning; Tumors; Cancer patients; Clinical features; Comparative analyzes; Demographic features; Learning techniques; Lung Cancer; Machine-learning; Forecasting

Lung cancer, one of the deadliest forms of cancer, can significantly improve patient survival rates by 60�70% if detected in its early stages. The prediction of lung cancer patient survival has grown to be a popular area of research among medical and computer science experts. This study aims to predict the survival period of lung cancer patients using 12 demographic and clinical features. This is achieved through a comparative analysis between traditional machine learning and deep learning techniques, deviating from previous studies that primarily used CT or X-ray images. The dataset included 10,001 lung cancer patients, and the data attributes involved gender, age, race, T (tumor size), M (tumor dissemination to other organs), N (lymph node involvement), Chemo, DX-Bone, DX-Brain, DX-Liver, DX-Lung, and survival months. Six supervised machine learning and deep learning techniques were applied, including logistic-regression (Logistic), Bayes classifier (BayesNet), lazy-classifier (LWL), meta-classifier (AttributeSelectedClassifier (ASC)), rule-learner (OneR), decision-tree (J48), and deep neural network (DNN). The findings suggest that DNN surpassed the performance of the six traditional machine learning models in accurately predicting the survival duration of lung cancer patients, achieving an accuracy rate of 88.58%. This evidence is thought to assist healthcare experts in cost management and timely treatment provision. � 2023, The Author(s), under exclusive licence to Springer Science+Business Media, LLC, part of Springer Nature.