Deep Neuro-Fuzzy System has been successfully employed in various applications. But, the model faces two issues: (i) dataset with many features exponentially increases the fuzzy rule-base, (ii) parameters in the fuzzy rule-base are optimized using the gradient descent approach, which has the drawback of local minima. Therefore, this study aims on improving the model's accuracy by proposing Arithmetic Optimization Algorithm. The outcomes using the Arithmetic Optimization Algorithm for feature selection have not only reduced the burden of implementing a huge dataset, but the Arithmetic Optimization-based deep neuro-fuzzy system has outperformed with 95.14 accuracy compared to the standard method with 94.52. © 2022 Elsevier B.V.