Comparison of Learning Methods Used in Data Mining Under Different Conditions

In this study, it is tried to determine the level of reliability and validity of data mining and machine learning approach in the field of education and the results obtained based on these algorithms. In the study according to the 2015 PISA average of Turkey in terms of scientific literacy, students were classified as failing and successful using different learning methods and the reliability and validity measures of the results obtained at this stage were examined. The best results in terms of the correct classification rate, kappa statistic, square root error and relative square root error values were obtained from Random Forest method besides this ridge logistic regression, logistic model and Hoefding tree methods were determined as the most successful other methods. In the case that the whole data set is separated as a training and test data set without using the cross validation method, it was found that the Random Tree and J.48 methods had the highest error values while the logistic model, Random Forest and Ridge Regression methods gave the lowest error values in different sized test data. It is concluded that the error values obtained by Ridge regression, Random forest and Logistic model are quite consistent in different percentage test data. It was determined that the measurement results obtained with the help of different methods did not divide the data set as test and training data, but if we train both the learning method and the same data set through the same data set, Random tree and J.48 learning methods had a higher accurate classification rate than the real performances.