The prediction of the voice traffic and their accurate model is a challenge for telecom provider. It is important for the quality & service analysis of mobile network. The study in this paper concern predicting the voice traffic (Erlang) of mobile network at busy hour (5 to 7 pm) of a day which is nonlinear & dynamic. This voice traffic also depends on many other non-linear &dynamic parameter which best predict for voice traffic. The data is collected from quality & service report (QOS) of telecom service provider. Correlation analysis is used to select proper input variables from QOS report. Feed forward back propagation algorithm is proposed to make traffic prediction at busy hours on daily basis.