Predicting Math performance of children with special needs based on serious game

Predicting and classifying student's performance using data mining techniques have been gaining an enormous amount of attention from researchers and practitioners. However, the use of games for the classification of student's ability level is still slightly. This study focuses on identification of important factors for determining student level performance on Math. The best classification algorithm is observed as part of intelligent game development research for assessment of children with special needs. The real dataset from randomly selected of elementary school is taken to construct a dataset. About 135 normal students and 25 children with special needs played the game and did a manual test. Our study shows that the age, gender, grade, and mark of each level became important factors in determining the level of math skill for the normal student. However age, gender, and grade don't have a correlation with math level of children with special needs. Six classification methods, Naive Bayes, Multilayer Perceptron (MLP), SMO, Decision Table, JRip, and J48, were performed to predict math skill performance level of normal students and children with special needs. JRip with 10 fold cross validation gives the highest percentage of accuracy of 64.12.