Multi step prediction is a complex task that has attracted increasing interest in recent years. The contribution in this work is the development of nonlinear neural network models for the purpose of building multi step Prediction of Internet Bandwidth i. e. bits per second transmission record of server. It is observed that such problems exhibit a rich chaotic behavior and also leads to strange attractor. . This paper compares the performance of four neural network configurations namely a Multilayer Perceptron (MLP) , generalized feed forward network(GFF) , Self organized feature map (SOFM), and the Jorden –Elmen network with regards to various performance measures Mean square error (M. S. E. ),Normalized mean square error (N. M. S. E) and regression (r) . The standard back propagation algorithm with momentum term has been used for all the models. There are various parameters like number of processing elements, step size, momentum value in hidden layer, in output layer the various transfer functions like tanh, sigmoid, linear-tan-h and linear sigmoid, different error norms L1,L2 ,Lp to L infinity, Epochs variations and different combination of training and testing samples are exhaustively experimented for obtaining the proposed robust model for long term as well as short step ahead prediction.

1. Barron, A R, Universal approximation bounds for superposition's of a sigmoid function, IEEET Transactions on information Theory,39,pp930-945,1993.
2. Juditsky , A Hjalmarson, H, Benveniste, A, Delyon , B,Ljung , L, Sjoberg, J & Zang, Q,Nonlinear black-box models in system identification: Mathematical foundation, Automatica,vol 31 , no 12,pp 1725,1995.
3. De Varies B & J C Principe, The gamma model-A new neural model for temporal processing, Neural Network, vol5,pp565,1992.
4. Sanjay V Dudul Identification of a Liquied of a Liquid Saturated Steam Heat Exchanger Using Focused Time Lagged Recurrent Neural Network Model, IETE Journal of Research Vol 53 No1,January-February2007,pp69-83.
5. K. V. Rajaram, M. A. Faruqi, Development of intelligent system, for predicting sticker induced breakouts in continuous casting based on artificial neural networks, Final Report ME/1998/ TISCO/2, Indian Institute of Technology, Kharagpur, India (September 1998).
6. Naraendra K S & Parthasarathy , K, Identification and control of dynamic systems using neural networks,IEEE Trans on Neural Networks,vol 1 , no 1 , pp4-27,1992.
7. Quin Waibel ,A, T Hanazana, G Hinton K Shikano & K J Lang, phoneme recognition using time delay neural Transactions on Acoustics , speech , and signal processing vol ASSP-37,pp328,1989.
8. Jose C Principe , Neil R Euliano & W Curt Lefebvre, Neural and Adaptive Systems –Fundamentals Through Simulation, John-Wiley & Sons,Inc,2000.
9. S Z, Su H T & Mc Avoy T J, Comparison of Four Neural Net Learning Methods for Dynamic System Identification , IEEE Transactions on Neural Networks, vol 3 ,no 1, pp 122-30, 1992. [10 ] Kohonen Springer Verlag, "self organizing feature map" Berlin, New York 1995.
10. S. Haykin, Neural Networks: A Comprehensive Foundation, second ed. , Prentice-Hall, Englewood Cliffs, NJ, 1999.
11. S. K. Chand, Modeling of complex systems from time series data: a computational intelligent approach, Ph. D. thesis, Indian Insti¬tute of Technology, Kharagpur, India (January 2000).
12. J. C. Sprott, Chaos and Time-series Analysis, Oxford University.
13. Press, NY, 2003
14. K. Funahashi: On the Approxima¬tion Realization of Continuous Map¬ping by Neural Networks, Neural Networks. vol. 2, No. 3, 1989
15. K. Doya, S. Yoshizawa: Memoriz¬ing oscillatory patterns in the analog neural network, proceedings of the 3rd IJCNN I, 27-32, 1989
16. M. Casdagli:- Non-linear predictions in chaotic time series physical D,35 335_359,1989.

Chaotic Multi Step Prediction Cross Validation.