The 5G wireless networks support diverse use-cases particularly ultra-reliable low-latency communications (URLLC). One of the key challenges in supporting URLLC services is to enhance the performance of the random access procedure to guarantee the stringent latency requirements. This is not only challenging for URLLC services but any delay sensitive services like voice over LTE (VoLTE) or voice over New Radio (VoNR). The access and mobility procedures rely on the random access procedure. Enhancing this procedure using artificial intelligence can thus support even more stringent latency requirements. In this paper, we present an experimental study aiming at performance evaluation of the access and mobility procedures based on an experimentation and data collection from the Monroe platform. We study the main causes of the delay induced to the access and mobility procedures, and evaluate machine learning based techniques to classify different procedures in terms of experienced delay and failure. Such results take step towards enabling the User Equipment (UE) to take appropriate actions for coping with predicted sub-optimal access or mobility procedures.