Detecting semantic code clones is crucial in software engineering to identify code fragments with similar functionality but different syntax. This process is vital for maintaining code quality and improving software maintainability. With recent advancements in machine learning and high computational processing, researchers increasingly utilize these sophisticated techniques for code clone detection. This paper presents a semantic code clone detector that utilizes machine learning algorithms on top of the low-level abstract compiled code intermediate representation known as Baf. The proposed technique has been implemented and assessed using real-world code clone pairs from the BigCloneBench dataset written in Java. Multiple experiments have been conducted, and the results have shown that the proposed technique outperforms several baseline methods in detecting semantic clones regarding recall and an F1 score.