Nested logistic regression is a very popular method in several areas, because it is easy to interpret the estimated result and it has a good predicted performance. However, when applying multiclass logistic regression to rare event data, which has the majority of class labels are some class label, all class labels are estimated as the “majority class”. In binary logistic regression, the cost-sensitive method was proposed to improve a prediction performance. In this study, we developed a prediction performance of nested logistic regression applied to multiclass rare event data. Then, we proposed a modified objective function to estimate parameters for multiclass rare event data using the classification cost. Experimental results on real-world data show that the proposed method is efficient for multiclass rare event data.