Comparison of Adaptive Boosting and Bootstrap Aggregating Performance to Improve the Prediction of Bank Telemarketing

Background: Telemarketing is an effective marketing strategy lately, because it allows long-distance interaction making it easier for marketing promotion management to market their products. But sometimes with incessant phone calls to clients that are less potential to cause inconvenience, so we need predictions that produce good probabilities so that it can be the basis for making decisions about how many potential clients can be contacted which results in time and costs can be minimized, telephone calls can be more effective, client stress and intrusion will be reduced. strong. Method: This study will compare the classification performance of Bank Marketing datasets from the UCI Machine Learning Repository using data mining with the Adaboost and Bagging ensemble approach, base algorithm using J48 Weka, and Wrapper subset evaluation feature selection techniques and previously data balancing was performed on the dataset, where the expected results can be known the best ensemble method that produces the best performance of both. Results: In the Bagging experiment, the best performance of Adaboost and J48 with an accuracy rate of 86.6%, Adaboost 83.5% and J48 of 85.9%Conclusion: The conclusion obtained from this study that the use of data balancing and feature selection techniques can help improve classification performance, Bagging is the best ensemble algorithm from this study, while for Adaboost is not productive for this study because the basic algorithm used is a strong learner where Adaboost has Weaknesses to improve strong basic algorithm.