Comparative Analysis of PCA and Autoencoder Feature Extraction Methods for Breast Cancer Detection Using MLP Neural Network

Breast cancer is one of the most common and aggressive cancers, making urgent and accurate diagnosis crucial for patient survival and treatment. With increasing patient data and advances in data mining, developing intelligent models to quickly and accurately distinguish benign from malignant tumors is essential. Data-driven models in medical applications must combine high accuracy with simple structures and few parameters to ensure reliable clinical use. This study proposes a model that achieves high accuracy with fewer parameters for breast cancer classification. During preprocessing, original features are mapped to a new space using linear reduction with Principal Component Analysis (PCA) and nonlinear reduction with an Autoencoder. These features, along with the raw data, are then used to train a multilayer perceptron (MLP) neural network for cancer classification. The methods are evaluated on the standard Breast Cancer Wisconsin dataset from the UCI Machine Learning Repository. K-Fold cross-validation is applied with metrics including Accuracy, Precision, Recall, and F-Score. Results show that the Autoencoder-based classifier not only reduces parameters and computational cost but also shows superior or competitive performance compared to classifiers trained on raw or PCA-based features, demonstrating its effectiveness in breast cancer detection.

کلیدواژه ها:

Classification ، MLP ، PCA ، Autoencoder ، Breast Cancer