The proposed DSL method shows promising performance in achieving a satisfactory AR and reducing runtime. However, it struggles in more complex scenarios where users have diverse distributions or in dynamic environments where users may change locations frequently. A potential solution is to train the DSL model on larger, more generalized datasets. However, this approach is impractical due to the immense computational resources required to generate such datasets, and even then, it may fail to encompass all possible scenarios. These limitations of DSL motivate us to explore RL as a more effective optimization strategy within ML.