Major challenges in current machine learning trend applications are scalability, collaborative filtering, stability, and accuracy. To overcome these types of challenges, we utilized the deep Q-learning algorithm. This is a model-free reinforcement learning algorithm, which is mainly addressed to overcome issues like collaborative filtering and will give an accuracy of 80–95% when compared to the other algorithms. To implement this algorithm, we designed an emotion-based music recommendation system. It allows users to search for music that is appropriate for their present emotional state. Deep Q-learning has been shown to be effective at learning to perform a variety of tasks, including emotion-based music recommendation. To accommodate extensive and current data, our system acquires information via APIs as opposed to relying solely on datasets. The proposed system is tailored to the user’s expressions, and this will frequently be used to classify these expressions into a variety of emotions such as happy, sad, and normal. Notably, deep Q-learning has proven its effectiveness in mastering a diverse array of tasks, particularly showcasing its proficiency in the domain of emotion-based music recommendation.