Age-Adaptive Lung Cancer Prediction Using Ensemble Learning Methods

Lung cancer, a leading cause of cancer-related deaths globally, is caused predominantly by smoking. This study compares machine learning (ML) and ensemble learning (EL) models to predict lung cancer across different age groups, high-lighting an age-adaptive approach. Data preprocessing involved handling null values, deduplication, and encoding categorical variables. Key lung cancer risk factors were selected using chi-square tests, and data was divided into young-aged, middle-aged and old-aged cohorts, with standardized scaling applied for computational efficiency. The data point from the young cohort was excluded due to the presence of a singular data instance. Furthermore, due to segregation of cohorts, the data left was imbalanced. To up-sample the imbalanced data, Synthetic Miniority Oversampling Technique (SMOTE) was used. For each machine learning and ensemble learning model, an optimal set of parameters was determined using hyperparameter tuining tool from scikit-learn library - GridSearchCV. Given the limited quantity of data, the models with these selected parameters were then trained using K-Fold Cross Validation to ensure robust performance. The results show that for middle-aged individuals, K-Nearest Neighbors (KNN) and Random Forests (RF) achieved top accuracies of 92.22% and 92.78% respectively, while for old-aged cohorts, Gaussian Naive Bayes (NB) and RF yielded 95.66% and 96.98 %. This age-specific analysis emphasizes the RF model's robustness in predicting lung cancer risk across age brackets.