Advancements in Breast Cancer Diagnosis: Integrating Classifier Algorithms, Neural Network and Ensemble Learning with PCA, VIF for Feature Selection and Dimensionality Reduction

Breast cancer's global prevalence highlights the need for the development of precise and reliable diagnostic tools. The objective of this study is to contribute to the growing body of knowledge in breast cancer diagnosis, highlighting the potential of a range of classifier algorithms, soft and hard voting ensemble approaches, and neural networks as potent tools in medical applications. These models were utilized to assess the Wisconsin Breast Cancer dataset obtained from UCI Machine Learning repository, consisting of 569 samples and 30 features. Besides, we utilized Principal Component Analysis (PCA) and Variance Inflation Factors (VIF) techniques to perform feature selection and dimensionality reduction on the standardized and original features respectively. After conducting PCA analysis, a variety of classifier models, including k-nearest neighbors (KNN), Lo-gistic Regression (LR), Decision Tree (DT), LightGBM (LGBM), XGBoost (XGB), Random Forest (RF), and Naive Bayes (NB), were employed. Moreover, after the VIF analysis, these classifier models and a Neural Network (NN) model were put into action. Subsequently, the best three and best five classifier algorithms were determined using accuracy metrics, then both soft and hard voting ensemble were executed on these algorithms. The neural network (NN) model underwent training for 500 epochs since beyond that point, the loss curves displayed nearly constant values. This model (NN) were compiled with "adam" optimizer along with binary crossentropy as loss function. We observed our ensemble strategies demonstrated superior performance in accuracy compared to all existing methods.