Abstract: The arc initiation process of P-GMAW is prone to instability, which can significantly affect the accuracy of arc sensing and weld seam tracking. In response to this issue, a study was conductedon on the causes of unstable arc initiation in oscillating arc narrow gap P-GMAW, revealing the significant impact of wire feed speed on the stability of the arc initiation process. By comparing and analyzing arc images and electrical signal characteristics, key signal features representing arc stability are extracted. To reduce redundancy and overfitting, a maximum likelihood estimation method is employed to select and extract 8 variables, which are then fused using principal component analysis to extract the top two components with the highest variance contribution. Based on the factor loading, it is found that the pulse peak stage is the phase where the arc is more prone to instability compared to the droplet transition stage and the baseline stage. By combining the extracted principal component variables with a binary logistic regression model, an arc stability discrimination model for the arc initiation process is established. The optimal threshold model was obtained through the ROC curve. Experimental validation shows that the model achieves an accuracy rate of over 80% in pulse stability discrimination, indicating good discriminatory performance. This model holds certain application value in improving the accuracy of narrow gap high-low tracking and ensuring welding quality.