DEVELOPMENT FORECASTING MODEL FOR INTERNET DATA TRANSFER USING NEURAL NETWORK TECHNIQUE

Abstract

The objectives of this study were: (1) to study the forecasting model of Internet traffic volume using neural network techniques, (2) to develop the forecasting model for Internet traffic using neural network techniques, and (3) to evaluate an effectiveness of forecasting models by neural network techniques.

The researcher used the 12 data sets of National Telecom Public Company Limited between Januarys to December 2021 and divided each data set into 2 sets. The first set is for learning forecast models and the second set is for testing predictive models with a ratio of 70 to 30. The input node (n) for a backward propagation neural network has 5 nodes and the output has 1 node and the neural network uses the Levenberg-Marquardt algorithm to train the network for tuning weights, bias values, and the number of nodes in different hidden layers to design a neural network that makes the lowest of the Mean Absolute Percentage Error (MAPE).

The results showed that (1) the forecasting model was a multi-layer perceptron neural network technique suitability for forecasting Internet traffic and (2) the model was developed using the technique of the Levenberg-Marquardt algorithm to find the optimum number of nodes in the hidden layer equal to 2 times of input data and the result of the performance evaluation found that (3) the forecasting model has been developed to forecast Internet traffic for 1n, 2n, 3n, 4n and 5n  models have the MAPE are 3.59, 3.01, 4.02, 4.08, and 5.95 respectively which the 2n model was the model that gives the lowest value of MAPE.