SEMeL-LR: An improvised modeling approach using a meta-learning algorithm to classify breast cancer

Abstract

In the last two decades, cancer has continued to be prone to a larger extent in females globally. In this regard, early prevention and treatment can help individuals to anticipate unnecessary deaths. At present, many advanced machine learning (ML) techniques are widely used but are still unable to diagnose breast cancer completely. Hence, in this research, a novel Stacking Ensemble Meta-Learning (SEMeL-LR) technique has been proposed that combines the predicted results of base ML models to get the best outcome with five different phases: (i) Collect Wisconsin Breast Cancer Database (WBCD) from the Kaggle repository, (ii) training original data at primary level (i.e. L0), (iii) predict with ML models at L0, (iv) build MeL model, and (v) final prediction using MeL at L1. First off, this study provides a significant contribution toward a deeper comprehension of the MeL technique, and secondly, is to compare our proposed model with five more common ML models to find the best predictive model. The experiment was carried out using Python 3.8.8 programming language on the Jupyter Notebook 6.4.3 platform, where we found the proposed model resulted in an accuracy of 98.59%, precision of 99.00%, recall of 99.00%, f1-score of 98.00%, true positive rates of 99.00%, and false positive rates of 1.00%, which was better than many well-known state-of-the-art ML models. Furthermore, we have implemented GridSearchCV () method for parameter tuning that takes 4 attributes with multiple values inside ‘params’ and found the best parameters as ‘penalty = [11, 12]’, ‘solver = ‘liblinear’, ‘c = 1438.45’ and max_iter = 100′ when grid.best_params_ had called. In the end, we tried with new sample values and found our proposed model had successfully classified the class as ‘1’, which represents a patient affected with breast cancer disease.