At level 3 of autonomous driving, the driver has to take over driving when the system requires. In automatic driving, the arousal level tends to decrease. Drowsiness or less arousal is the leading cause of car accidents. For safety, it is necessary to increase the arousal level before driving. Moreover, due to the emotional state effect on the driving performance, it is important to consider comfort while improving the driver's arousal level. Previous studies proposed the comfortable arousal model based on physiological signals to evaluate arousal and comfort during autonomous driving. However, the accuracy evaluation using this model has not been sufficiently performed. This study aims to construct a more accurate and reliable comfortable arousal model. We explore various physiological indexes and calculate feature importance using the random forest method to achieve our goal. Then we compare and validate the evaluation accuracy with the subjective evaluation score against the previous comfortable model proposed. The result shows that the proposed method has more accurate than the methods of the previous method. However, the improved accuracy is still not very high, so we need to consider creating a comfortable arousal model.