Fig. 3. SVM modeling predicts goal-directed and habitual reward seeking using external globus pallidus (GPe) arkypallidal neuron Ca²⁺ dynamics.

a Averaged receiver operator characteristic (ROC) curves for prediction of goal-directed or habitual reward-seeking around behavioral events (nose-poke, magazine entry, and magazine exit; −2 – 2 s input range). b Comparison of accuracy, sensitivity, and specificity to random chance among each behavioral event. Two-tailed one sample t tests were used for (b). For accuracy NP versus 50%, t = 116.1, p < 0.0001, for accuracy M_entry versus 50%, t = 136.4, p < 0.0001, for accuracy M_exit versus 50%, t = 105. 7, p < 0.0001. For sensitivity NP versus 50%, t = 109.1, p < 0.0001, for sensitivity M_entry versus 50%, t = 126.9, p < 0.0001, for sensitivity M_exit versus 50%, t = 101.4, p < 0.0001. For specificity NP versus 50%, t = 104.8, p < 0.0001, for specificity M_entry versus 50%, t = 118.9, p < 0.0001, for specificity M_exit versus 50%, t = 94.35, p < 0.0001. *p < 0.05 for comparison to random chance for accuracy, sensitivity, and specificity. See Supplementary Table 1 for AUC values for additional time input ranges. See Supplementary Table 2 for full statistical information. Source data are provided as a Source Data file.