BREAST CANCER PREDICTION FROM GENE EXPRESSION DATA USING RECURRENT NEURAL NETWORKS

Gene expression data holds significant potential for identifying biomarkers and predicting the progression of breast cancer. Despite advancements in machine learning, accurately predicting breast cancer from gene expression data remains a challenge due to high-dimensionality, noise, and feature correlation in datasets. This study proposes a hybrid Recurrent Neural Network (RNN) to enhance prediction accuracy. The RNN combines convolutional layers for feature extraction with recurrent layers to capture sequential dependencies inherent in gene expression data. The method begins by preprocessing the gene expression dataset through normalization and feature selection techniques to reduce dimensionality. The RNN model incorporates convolutional layers to extract spatial patterns and long short-term memory (LSTM) layers to capture temporal dependencies. Batch normalization, dropout, and adaptive optimizers are applied to prevent overfitting and improve convergence. Experimental evaluation using a publicly available breast cancer gene expression dataset demonstrates that the proposed RNN model outperforms existing methods, achieving an accuracy of 97.5%. Comparisons with Support Vector Machine (SVM), Random Forest (RF), and Deep Neural Networks (DNN) highlight the RNN''''s superiority in handling complex, high-dimensional data.