Preliminary Analysis to Investigate Accuracy of Data Mining for Childhood Obesity and Overweight Predictions

摘要

Childhood obesity is a very worrying global epidemic and the Malaysian children have shown alarming statistics. Therefore, obesity and overweight predictions at an early age are important. This paper presents performances of eleven data mining techniques, that are sensitivity, specificityand accuracy tested using 320 Malaysian children datasets that were collected. The data mining techniques are decision tree, Support Vector Machine, Neural Networks, Discriminant Analysis, K -means Clustering, Regression and Na?ve Bayes. The results indicated that the Classificationand Regression Tree has shown high specificity in normal and obesity predictions, while the Naive Bayes has shown high sensitivity in overweight and obesity predictions. Meanwhile, other techniques have adequate or poor accuracy. Overall, the data mining techniques accuracy can be improvised.Previous studies also indicated that the data mining techniques have limited prediction accuracy. Therefore, based on analysis, the data mining techniques can be enhanced to address the issue of low prediction accuracy for childhood obesity and overweight predictions.