Comparison of three data mining algorithms for potential 4G customers prediction

The size and number of telecom databases are growing quickly but most of the data has not been analyzed for revealing the hidden and valuable intellectual. Models developed from data mining techniques are useful for telecom to make right prediction. The dataset contains one million customers from a telecom company. We implement data mining techniques, i.e., AdaboostM1 (ABM) algorithm, Naïve Bayes (NB) algorithm, Local Outlier Factor (LOF) algorithm to develop the predictive models. This paper studies the application of data mining techniques to develop 4G customer predictive models and compares three models on our dataset through precision, recall, and cumulative recall curve. The result is that precision of ABM, NB and LOF are 0.6016, 0.6735 and 0.3844. From the aspects of cumulative recall curve NB algorithm also is the best one.