Non-contact mapping (NCM) is criticised primarily due to reduced accuracy for atrial walls located > 4 mm from the catheter. Despite these limitations, NCM uniquely provides real-time simultaneous sensing of global atrial activation. Alternative technologies, such as basket catheters, often encounter issues including poor wall contact and suboptimal resolutions. Rotor/re-entry mapping showing great success in some centres, driving industry and clinical interest towards sequential rotor mapping using catheters like the PentaRay and HD-Grid. However, such sequential mapping inherently assumes signal stability. Atrial fibrillation (AF) represents one of the most complex atrial activation patterns, therefore, unstable. Our prior studies indicate that dominant frequency (DF) lacks spatiotemporal stabilities, and capturing stable DF requires at least 84 seconds, whereas accurate phase singularity (PS) density mapping can be achieved within 18 seconds. We conducted DF-guided ablation utilising a near-real-time user interface to target clusters of high DF regions, achieving termination of AF in 4 out of 10 cases even before pulmonary vein isolation (PVI). Further analysis revealed recurrence of HDF regions is more strongly correlated with stability rather than the highest DF sites. With data labelled by real AF cycle length changes following targeted ablation, machine learning models datasets achieved promising accuracies up to 70%, employing Recurrence Quantification Analysis (RQA) features and advanced wavelet clustering algorithms. Despite the discontinuation of EnSite arrays and reduced utilisation of Acutus mapping systems, early analysis highlights inherent challenges but underscores significant potential for machine learning algorithms when combined with clinically annotated datasets to substantially enhance AF mapping precision. Our simultaneous recordings from EnSite arrays and PentaRay catheters, can be used to bridge contact and non-contact mapping methods, enabling furthering understanding and adaptation of machine learning algorithms to contact mapping platforms. Integrating these concurrent mapping methods with advanced computational techniques holds notable promise for driver identification.