Early diagnosis and treatment of cancer are important steps in improving patient outcomes. In recent years, deep learning models have significantly assisted the medical field, improving the diagnostic and treatment capabilities of clinicians. However, individual models often produce inconsistent results and are sensitive to variations in input data, which can affect the classification performance of machine learning systems. In this paper, we propose the RIDOPT-Net model, which is designed to improve classification performance by integrating multiple machine learning models and optimizing their weights to reduce the classification error rate. The input data undergoes pre-processing to remove noise, normalize values, and enhance image quality before training. Based on each model’s performance, the weights are automatically optimized using Optuna, enabling more effective ensemble integration. The proposed approach achieves a classification accuracy of 97.68% on the test dataset, demonstrating improved robustness and reliability. This study contributes to the development of a reliable and efficient breast cancer early detection system using intelligent systems and advanced data processing techniques in important healthcare applications.