To ensure the integrity and safety of online interactions, many fake profiles on social media are often created by harmful bots. Despite the tactics already employed, their efficacy is frequently compromised by flaws like the reliance on static profile traits and the inability to consider changing behavioral patterns. These flaws call for a more reliable and flexible method to identify bot-based fake profiles precisely. This paper offers a novel model that uses Deep Dyna-Q Networks (DDQN) to fill this gap, improving the precision and speed of bogus profile identification. In contrast to conventional approaches, the proposed model takes a closer look at an account’s activity, followers, and following in addition to its profile traits. This method considers the number and quality of these parameters and, as a result, can spot subtle patterns that may point to a fake account for real-time scenarios. The suggested methodology uses multidomain feature analysis, making understanding intricate, linked behavioral patterns easier. This in-depth investigation makes it easier to spot the abnormalities in activity, correlations in followers and followings, and discrepancies in profile information that are direct signs of fake profiles created by bots. The study’s findings show that the suggested model performs better than others. In addition to a decrease in detection delay, there is an improvement in precision, accuracy, recall, A.U.C., sensitivity, and specificity levels. These enhancements demonstrate the model’s potency and highlight the possibility of using cutting-edge deep learning algorithms to address the growing problem of fake profiles on social media sites.