Even by using the data mining, many weaknesses still existed in childhood obesity prediction and it is still far from achieving perfect prediction. This paper studies previous steps involved in childhood obesity prediction using different data mining techniques and proposed hybrid approaches to improve the accuracy of the prediction. The steps taken in this study were a review of childhood obesity, data collections, data cleaning and preprocessing, implementation of the hybrid approach, and evaluation of the proposed approach. The hybrid approach consists of the classification and regression tree, Naïve Bayes, mean value identification and Euclidean distances classification. The results from the evaluation have shown that the proposed approach has 60% sensitivity for childhood obesity prediction and 95% sensitivity for childhood overweight prediction.