Improving diagnostic accuracy for breast cancer using prediction-based approaches

Abstract

The objective of this study is to improve prediction outcome of breast cancer patients employing Machine Learning techniques so to be able to accurately classify between Benign or Malignant Tumor. The dataset taken for this experiment is an inclusion of extracted features of breast cancer patient cells and normal person cells that are extracted from digitized images of FNA (Fine-needle aspiration) tests performed on breast lumps. The dataset was exposed to Machine Learning models namely Support Vector Machine, Decision Tree, Logistic Regression, K- Nearest Neighbor, Naive Bayes, Random Forest and Neural Network based algorithm- Multilayer Perceptron to analyze the prediction results. The obtained results were also compared with ensemble- based learning techniques such as Gradient Boost, XGBoost and Adaboost classifiers to find the best performing algorithm. Further, this study aims to showcase to the clinicians the methodology of interpretation via Machine Learning and that it's routinely usage would certainly be beneficial to predict outcomes. The long-term goal of this type of study expects a slow and gradual realization of the importance of accurate tumor detection via Machine Learning models, as early detection of breast cancer can greatly improve prognosis and survival chances by promoting clinical treatment to patients as soon as possible.