The increased flexibility and density of spectrum access in 5G NR have made jamming detection a critical research area. To detect coexisting jamming and subtle interference that can affect legitimate communications performance, we introduce machine learning (ML)-assisted Bayesian Inference for jamming detection methodologies. Our methodology leverages cross-layer critical signaling data collected on a 5G NR Non-Standalone (NSA) testbed via supervised learning models, and are further assessed, calibrated, and revealed using Bayesian Network Model (BNM)-based inference. The models can operate on both instantaneous and sequential time-series data samples, achieving an Area under Curve (AUC) in the range of 0.947 to 1 for instantaneous models and between 0.933 to 1 for sequential models including the echo state network (ESN) from the reservoir computing (RC) family, for jamming scenarios spanning multiple frequency bands and power levels. Our approach not only serves as a validation method and a resilience enhancement tool for ML-based jamming detection, but also enables root cause identification for any observed performance degradation. Our proof-of-concept is successful in addressing 72.2\% of the erroneous predictions in sequential models caused by insufficient data samples collected in the observation period, demonstrating its applicability in 5G NR and Beyond-5G (B5G) network infrastructure and user devices.