EHA-LNN: Optimized light gradient-boosting machine enabled neural network for cancer detection using mammography

Breast cancer is the most fatal illness among adult females, which should be detected earlier to reduce mortality and increase the chance of complete recovery. Several research works are formulated using deep learning techniques for early detection, but those techniques show less accuracy due to the inefficiency of classifiers. Also, the main challenges faced by the previous models were over-fitting, interpretability, and increased false positives that limited their performance. Therefore, this research proposes an Enhanced Harmony Search Algorithm based Light Gradient-Boosting Machine-Enabled Neural Network (EHA-LNN) model using mammography images for effective breast cancer detection. Specifically, the EHA-LNN model combines light GBM and neural network to minimize the information loss during computation and thereby reduces false negatives. Also, the EHA-LNN model effectively handles large datasets of mammography images, which reduces over-fitting issues and improves interpretability for better detection. Utilization of EHA optimization benefits to gain optimal results by efficient parameter tuning and also increases the convergence speed. Moreover, the EHA-LNN model achieved 98.63% accuracy, 98.99% sensitivity, and 98.27% specificity compared to other conventional methods.