The goal of Content-Based Image Retrieval (CBIR) is to compare a query image with similar images from a large dataset. Typically, the ranking of retrieved photos is based on how similar the representative features of the query image and the dataset images are. Machine learning (ML) methods have been investigated as a practical approach to decrease the semantic gap. This research proposes a novel technique that utilizes evolutionary machine learning in CBIR. The input image is processed and classified using a Kernelized Radial Basis Auto-Encoder Function Neural Network (Ker_RadBAEFNN). Then, the input neural network is optimized using reinforcement in CBIR. Experimental analysis is carried out in terms of accuracy, precision, recall, F-1 score, RMSE, and MAP for various input datasets. The suggested architecture has exceptional performance in feature learning without prior knowledge of the images, categorization, and optimization. The proposed method achieved 98% accuracy, 96% precision, 79% recall, a 66% F-1 score, a 69% RMSE, and a 59% MAP.