Towards Forecasting Revenue Collection Using Multi Layer Perceptron (MLP)

Royal Customs Department Malaysia (RCDM) has the main function to collect revenue through indirect taxes, instead of giving facilitation and encourage industrialization and trades according to current provisions by laws and regulations towards vision 'to be recognized respected and world class services'. RCDM was setting revenue collection target yearly along with Strategic Planning 2001-2005. The revenue collection target has become an important agenda for every State Director of Customs through nationwide. Every state has to set their collection target based on the capability and tax resources, licensees and current performance. Forecasting revenue collection used statistical fundamental model has been used by RCDM by the year 2002 via deployment completed forecasting software that recognized as Forecast Pro Version 4.0. The significant of revenue forecasting is it provided a method to monitor collection performance and to take effective planning in order to ensure that revenue collection target can be achieved. This is as a result of the revenue collection performance was depends on various factors such as economics, politics, government policy and business environment that always been changing. This study is more on a new exploration technique using Artificial Neural Network (ANN) towards forecasting revenue collection of RCDM. The data sets were gathered from Monthly Revenue Return that provided and allowed to be used by Technique Division, RCDM, Alor Setar, Kedah. The data sets comprises of 1727 data that composed by 7 types of the duty and tax with non tax revenue from 159 successive month starting from 1st January, 1990 to 30 Mac, 2004. ANN with back propagation model, MLP has been used to train data sets in order to develop forecasting model revenue collection. Hopefully this study can be assist RCDM to develop a complete forecasting revenue collection tools using ANN for future enhancement. This study has proven the capability and reliability of ANN towards forecasting revenue collection. The results from BP model have proven the accuracy of forecasting is more than 92 percent. The ANN was found that can feed the data towards forecasting revenue collection. Thus, making is faster and easy to use. On the other hand, this study also adds more study domain related to current ANN applications for Faculty of Information Technology, Northern University of Malaysia.