Fig. 5

Heatmap and RCS of variables with sharp cutoffs using the SHAP method. Figure 5 describes the restricted cubic spline analysis of variables with obvious cutoff points and the heat map of these variables. In order to further observe the correlation between variables and predicted results, we performed RCS. Figure 5-a1 ~ 5-a4 show the nonlinear correlation between variables with obvious cutoff points and predicted results. The horizontal axis of the RCS series of graphs is the variable size, and the vertical axis is the Odds radio value between paroxysmal and persistent AF. Figure 5-a1 and 5-a2 show that LA and Hb have similar correlations with paroxysmal and persistent AF subtypes, which are protective factors at low values ​​and risk factors at high values. Figure 5-a3 shows that LVEF is a risk factor at low values ​​and a protective factor at high values. For these variables, we further show their heat maps. The charts shown in Fig. 5-b1 ~ b4 visualize 500 samples, with the X-axis representing the size of the variables and the Y-axis representing their impact on the results. The red area indicates a greater tendency to be diagnosed with persistent AF, and the blue area indicates a greater possibility of being diagnosed with paroxysmal AF. When there is a clear dividing point between the red and blue areas of a variable, it means that the variable may have a certain cutoff point, and the values ​​before and after this value point to completely different diagnoses