Abnormal telephone identification via an ensemble-based classification framework

Abnormal telephone that often appeared in our daily life has nearly crazed all people. Most of existing solutions to the problem cannot identify them in time with low efficiency and poor prediction. By cooperating with a telecom company, we have collected a certain amount of telecom user data. But, by analysis on the data, we found that it is a special classification problem in the case of small-scale data with noise samples. In this paper, we propose an ensemble-based classification framework, which first generates multiple training sets by partly resampling on the original data, then build classifiers for every of the generated training sets and finally combines the classifiers in the ensemble way. We conduct experiments on real-life data to evaluate the performance of our framework in comparison with some practical classification algorithms. The empirical result and analysis demonstrate that our framework can achieve a significant increase in accuracy for such difficult prediction task, especially when having noise samples.