Fibrosis is strongly linked with the progression of atrial fibrillation (AF), the most common sustained arrhythmia. Along with the overall fibrosis burden, its spatial distribution has been shown to play a key role in AF mechanisms. This study aims to predict AF inducibility using metrics of fibrosis heterogeneity and pulmonary veins (PVs) anatomy derived from patient MRI data. LGE-MRI scans from 20 patients were used to generate personalised left atrium (LA) models with three fibrosis levels, based on image intensity ratios. The patient-specific models were then used to simulate four AF scenarios per model, corresponding to four pacing used to initiate AF. From each LA model, metrics of fibrosis (e.g., entropy, variability, patch size) and PV anatomy (e.g., perimeter, area, sphericity) were extracted. A Random Forest (RF) classifier was trained (5-fold cross-validation, 80/20 split) to predict AF inducibility outcomes and extract the most important features leveraged in the classification. The RF achieved an average F1-score of 85.7% on the validation set. Feature importance analysis revealed that the RF model’s decision was most influenced by such features as fibrosis entropy and variability at 1.0–2.0 cm from the PV ostia. PV anatomy features showed lower predictive value. The study highlights the importance of fibrotic heterogeneity as a strong predictor of AF inducibility, surpassing traditional anatomical PV metrics. Such structural imaging features can be derived from patient imaging for non-invasive substrate characterisation and may inform personalised treatment strategies.