This research work has prioritized and optimized the risk classification levels by Hybrid deep belief neural networks. The Mayfly algorithm has been utilized with a fuzzy inference system for the best test case selection and prioritization. In this work, the input dataset is collected from the public bug repository available on Git hub for the fifteen Java projects. Then, risks are identified and classified from the bug database by Hybrid deep belief neural networks. Based on the risk levels, test cases are selected, prioritized, and reduced in volume by fuzzy inference system and Mayfly algorithm has been used to improve fault detection rates. The quality of the proposed risk-based testing approach and model are assured in terms of the APFD (Average percentage of fault detection) performance metric.