A Comparative Study of Statistical and Data Mining Algorithms for Prediction Performance

The aim of this study is to perform a comparison experiment between statistical and data mining modelling techniques. These techniques are statistical Logistic Regression, data mining Decision Tree and data mining Neural Network. The comparison will evaluate the performance of these prediction techniques in terms of measuring the overall prediction accuracy percentage agreement for each technique. The ratio of the binary values of the dependent variable in the training dataset and the population is used on the three techniques to find the effect of this ratio on the prediction performance. For a given data set, the results shows that the performance of the three techniques is comparable in general with small outperformance for the Neural Network. An affecting factor that makes the prediction accuracy varied is the dependent variable values distribution (distribution of “0”s and “1”s). It is seen that, for all of the three techniques, the overall prediction accuracy percentage agreement is high when the ratio of “0”s and “1”s is 3:1, whereas for the ratios 2:1 and 1:1 the performance is lower. KeywordsData Mining, Classification, Prediction Model, Statistical Logistic Regression, Neural Network,