5. Conclusions
In this paper, we discussed a few different data mining algorithms and statistical techniques to
identify the fraudulent online transactions. The results show that all the adopted data mining
approaches including NaiveBayes, AdaBoostM1, NaiveBayesMultinomial and
ClassificationViaRegression produce highly accurate predictions. However, the high Type II error rates
from these approaches indicate that those approaches cannot identify the fraud cases effectively.
Therefore, LDA is introduced to deal with the unbalanced data. However, for the non-fraud cases,
LDA produces less accurate results than the data mining approaches.