An interpretable decision-making process is vital for developing recommender systems. It should identify key items and relevant information to improve the recommendations. Deep learning models have shown promising performance, but simultaneously achieving explainability and efficiency is challenging. Existing explainable recommender systems interpret feature importance after training, which may not be able to determine crucial feature importance for enhancing performance. Additionally, the model’s learning paradigm depends on a single-objective function that ignores the captured information of user-item interaction. This study aims to develop a feature importance aware deep neural network model for explainable recommender systems, which facilitates improved predictive performance by considering the feature importance within the training procedure and generates a recommendation list based on the user’s preference. The proposed method minimizes the combination of model losses and introduces a penalty term that specifically targets and diminishes the detrimental impact of a particular feature on the solution’s efficacy. This strategy ensures the determination of the optimal solution by satisfying the explainable conditions imposed during the training framework. Extensive experiments on publicly accessible FilmTrust and MovieLens-100K datasets show notable recommendation performance (The source code is available at: https://github.com/PS297/ExplainableRS.git ).