Introduction: Several strategies have been used to target persistent atrial fibrillation (persAF) driver. However, no unique strategy has been proven effective in patients with persAF. Machine learning (ML) classifiers using features extracted from wavelet scattering transform (WST) might enhance the ablation outcomes. Methods: 51 high DF locations were ablated in 10 patients. 3206 non-contact electrograms (EGMs) were collected pre- and post-ablation using a balloon catheter. 1490 EGMs were labelled as positive ablation responses (AF termination or AF cycle length (AFCL) increased (≥10msec)), whereas 1716 EGMs were labelled as negative responses (AFCL increase (<10msec)) to catheter ablation. The WST technique was applied to extract features from EGMs after applying the QRST subtraction process. Several wavelet functions and dimensionality reduction methods were used. 10 ML classifiers were trained and tested by leaving EGMs of one patient out 10-fold cross-validation (CV). Results: The 10-fold CV overall accuracy, sensitivity, specificity, precision, F1_ score, AUROC, and balanced accuracy for the best scenario were using the Morlet function in WST, PCA, and the decision tree model with 78.91%, 81.21%, 76.92%, 75.34%, 78.17%, 0.76 and 79.07%. Conclusions: WST, with the help of PCA, played a significant role in predicting the responses of ablating the EGMs and their effect on AF termination and CL changes. The obtained results demonstrate the superiority of this method over our previous work.