Cancer disease is a category of diseases distinguished as an uncontrolled increase and extension of unnatural cells within the body, often caused by genetic mutations and various risk factors. Breast cancer (BC) stands as a common cancer forms. The early detection through timely examination and treatment greatly improves the chances of a successful outcome. To enhance early detection and improve treatment outcomes, a gene expression data set was used, but the curse of dimensionality appears when trying to analyze such data. So, it is important to filter this disturbance and lower the lengths of the microarray data, which is considered an essential step. First, the dataset was normalized using the min–max scalar approach. To work to reduce the discrepancy between different values in the data, which helps Binary Harris Hawks Optimization (BHHO) in selecting features efficiently. The suggested approach is assessed k-nearest neighbor (KNN), support vector machine (SVM), logistic regression (LR), and neural network (NN) is investigated. When contrasted to the most recent techniques, the suggested approach can accurately identify breast cancer using a selective genes. With just 31 features, it achieved a 96.67% accuracy in classification for the Van’t Veer dataset that is benchmarked.