This project involves data mining on mobile users’ survey data generated as part of the Marketing Research course at ISB. The goal is to predict the mobile carrier preference of a customer based on mobile survey data. The purpose of this prediction is to enable the Retail store to use a predictive model to offer a mobile carrier service to the Customer along with his phone purchase.
We have used myriad classification methods such as K-NN, Naïve Bayes and Classification Tree for the purpose of predicting the mobile carrier service. Our error rate was between 50-60% in the Validation and Test data. Such error rate in a prediction is obviously not acceptable for offering the service to the customer. In order to improve the performance of data mining application, we used an ensemble method, which
actually gave us much better success by improving the model’s predictive ability.
The ensemble method uses the outputs of all the classification methods and then does a majority vote among the predictions of each model to arrive upon the most common prediction. Surprisingly, the
predictive accuracy substantially improved from 55% to 43-44% with this method. Although, this accuracy would still not be sufficient to deploy this model to predict the mobile carrier service desired by Customer, a more representative sample would have ensure higher predictive ability.
In conclusion, we found that, along with the classification methods such as Naïve Bayes, the ensemble method is a strong tool for improving the predictive accuracy of model used for business forecasting.