Lung Cancer Prediction using Combination of Oversampling with Standard Random Forest Algorithm for Imbalanced Dataset

Lung cancer is a prominent kind of cancer that affects individuals of both genders. The early identification and treatment of lung cancer are of utmost importance. The objective is to compare several oversampling techniques using a dataset pertaining to lung cancer. The presence of datasets with imbalances in the context of supervised learning is widely recognized as an ongoing difficulty for conventional algorithms. This is mostly due to the fact that these algorithms are designed to effectively handle datasets with balanced class distributions, and as a result, they tend to exhibit worse performance when faced with imbalanced scenarios. The proposed oversampling methods including combinations of ADASYN with standard random forest, random over-sampler with standard random forest, and SMOTE with standard random forest are compared with existing RF, SVM, MLP, and GB algorithms to achieve a balanced class distribution. To evaluate the proposed method, Kaggle lung cancer dataset was used and implementation was done by using Python tool. The results indicate that the proposed ADASYN with SRF produced high performance with the accuracy score of 99% and precision and recall, F-scores, and ROC values are 99.56, 98.24, 99.11, and 98%.