Breast cancer diagnosis with machine learning algorithms and AutoML

Machine learning plays an important role in many fields including medicine. Especially in the field of healthcare, machine learning serves as an important tool to help diagnose and treat various diseases. One of the important applications of machine learning is the early detection of diseases such as breast cancer which is the leading cause of cancer death worldwide especially among women. Achieving high accuracy in the diagnosis of breast cancer is very important as it directly affects the effectiveness of treatment and general condition of patient. Medical data can be accurately analyzed with data analysis and different machine learning models. In this study the comparative performance of two different approaches to machine learning model development on the Wisconsin Breast Cancer (Diagnosis) dataset is investigated using various machine learning algorithms (Logistic Regression, Decision Tree, KNN, Naive Bayes, Support Vector Machine, Random Forest, Stochastic Gradient Descent, Adaboost, XGBoost, LightGBM, Artificial Neural Networks) with preprocessing, data augmentation, hyperparameter optimization and AutoML methods (TPOT, H2O, MLJAR) with data augmentation and preprocessing. Standardization, SMOTE and random undersampling and hyperparameter optimization generally improved model results. AutoML methods achieved higher accuracy and F1-score values than all machine learning algorithms. AutoML methods show the potential to be an effective tool for automating the model building process and optimizing the model.