MapReduce-based Backpropagation Neural Network over large scale mobile data

Large scale mobile data are generated continuously by multiple mobile devices in daily communications. Classification on such data possesses high significance for analyzing the behaviors of mobile customers. However, the natural properties of mobile data presents three non-trivial challenges: large data scale leads it difficult to keep both efficiency and accuracy; similar data increases the system load; and noise in the data set is also an important influence factor of the processing result. To resolve the above problems, this paper integrates conventional backpropagation neural network in the cloud computing environment. A MapReduce-based method called MBNN (i.e. MapReduce-based Backpropagation Neural Network) is proposed to process classifications on large-scale mobile data. It utilizes a diversity-based algorithm to decrease the computational loads. Moreover, the Adaboosting mechanism is introduced to further ameliorate the performance of classifications. Extensive experiments on gigabyte of realistic mobile data are performed on a cloud computing platform. And the results show that MBNN is characterized by superior efficiency, good scalability and anti-noise.