COMPUTATIONAL MODEL FOR REVENUE PREDICTON BASED ON BACKPROPAGATION NEURAL NETWORK (Case Study: Non-Tax Revenue Prediction for Indonesian Gov. Unit)

• Main Author: ANSHORI LUBIS - NIM 23216067 , FADLY
• Format: Theses
• Language: Indonesia
• Online Access: https://digilib.itb.ac.id/gdl/view/27073
• Institution: Institut Teknologi Bandung

<p align="justify">High quality economics planning proved by high level of accuracy between planning and realization data. As well in government revenue prediction, specifically for non-tax revenue case study, planning using qualitative interpretation with partial information rather than data analysis is one of the problems in existing process. Furthermore, the high influence of non-deterministic variables such as internal policy, political situation and global economic factor, requires the non-tax revenue planning has to using the approximation calculation method in it’s calculation. This paper proposes the artificial neural network (ANN) as one of machine learning method as a solution in calculate the approximation prediction value represented in a computational model. The analysis process focused on two objects, data partitioning (partition - non partition data set) and number of hidden neuron as one of variable in neural network algorithm (obtained from related researches formulation). Both combined and compared then to get the best accurate model based on least error prediction. The final evaluation result, the combination of data with government unit attribute partition with 11 hidden processing neurons resulting MSE = 0,00001551 selected as the best model to proposed for this case study. The average percentage of error prediction of planning against the realization can be optimized from 58,07% using previuous analysis to 44,43% by using proposed model. <p align="justify"> <br />