Causal inference aids in estimating treatment effects for informed decision-making. Over time, there has been an increasing focus on customizing and tailoring intervention services, leading to the emergence of adaptive interventions. These interventions implement personalized sequences of treatment options by employing decision rules that integrate participant data to formulate customized recommendations for intervention progression. However, causal inference in adaptive interventions faces several challenges, a key one being the distinction between correlation and causation. This requires robust study designs, examples include the Sequential Multiple Assignment Randomized Trial (SMART), which is resource-intensive and complex. In addition, SMART data often contain confounding variables that mask the true interplay between interventions and their outcomes. Another obstacle is the diversity of populations; adaptive interventions that are effective in one subgroup may be ineffective in another. Response-adaptive designs potentially reducing the likelihood of selecting the most appropriate treatment for each individual. In this context, we present Q-learning, an extension of regression analysis designed for situations where decisions are made sequentially regarding choices or intervention options available. Utilizing Q-learning involves applying it to assess the effectiveness of intervention options. Specifically, we utilize Q-learning alongside linear regression to approximate the optimal sequence of decision rules. This integration demonstrates how Q-learning, when combined with SMART data, enhances the refinement of decision rules, surpassing those originally embedded within the SMART framework. Eventually, we illustrate this methodology using the adaptive interventions for BPPV SMART (Reg. No. ChiCTR2100048603).