Towards Enhanced Breast Cancer Prediction: A Comparative Evaluation Study
Ahmed Zafran Bin Ahmed Sharizan, Mohsen Marjani, Dalia Abdulkareem Shafiq, Noor Zaman Jhanjhi, David A L Asirvatham, Komal Sharma · 2024
Breast cancer remains a foremost cause of mortality among and prevalent malignancy globally, constituting 30 % of all cancer diagnoses in women as of 2022. Timely detection of cancer before metastasis and accurate diagnosis of tumors as malignant poses a significant challenge in healthcare. Deploying accurate predictive models is crucial for facilitating life-saving early interventions. This study presents a comparative evaluation of Machine Learning (ML) classification algorithms for tumor type through the analysis of patient tumor structure quantification. Our research employed supervised ML algorithms that distinguish between malignant and benign tumors using tuning parameters to discern the optimal configurations for multiplying precision and minimizing the error rates. Additionally, a comparative analysis of K-Nearest Neighbors (KNN), Logistic Regression (LR), and Random Forest (RF) algorithms is carried out. The study's findings highlight the superiority of the RF Classification model, which attained an impressive 97 % accuracy, surpassing KNN and LR, with recorded accuracies of 96% and 94 %, respectively. Furthermore, performance metrics such as accuracy, precision, recall, f1-score, true and false positive and negative rates, and area under the Receiver Operating Curve (ROC) were applied for a comprehensive analysis of the algorithms. The results reveal improved accuracy compared to existing literature by 0.9 %. Although limited to a single dataset, our methodology shows promising results for real-world applications using analogous input parameters. Future research should aim for the application of advanced tuning encompassing Deep Learning and Neural Networks.