The expression “code smell” implies a sign of an issue in the nature of the source code. Various examinations have planned to recognize risky highlights in source code, at first zeroing in on measurement-based and heuristic-based approaches. As of late, in any case, there has been a shift toward using AI and deep learning (DL) methods for smell discovery. In spite of this, current calculations are still in the early advancement stages. Perceiving the difficulties of recognizing smells utilizing DL techniques, endeavors have been made by the two scholastics and programming engineers to address these hindrances. This work includes building and assessing new DL models for code smell discovery, with two models based upon stacked autoencoders utilizing a half-and-half engineering joining bidirectional long transient memory and convolutional brain network parts.