An Analysis Of Computational Learning Models For Quit Rent Revenue Estimation

• Main Author: Hamdan, Muhamad Hamiza
• Format: Thesis
• Language: English; English
• Published: 2017
• Subjects: Q Science (General); QA76 Computer software
• Online Access: http://eprints.utem.edu.my/id/eprint/20747/1/An%20Analysis%20Of%20Computational%20Learning%20Models%20For%20Quit%20Rent%20Revenue%20Estimation%20-%20Muhamad%20Hamiza%20Hamdan%20-%2024%20Pages.pdf; http://eprints.utem.edu.my/id/eprint/20747/2/An%20Effect%20Of%20Surface%20Roughness%20To%20Colour%20Sensor%20Detection%20Range%20In%20Ambient%20Temperature.pdf; http://eprints.utem.edu.my/id/eprint/20747/; http://libraryopac.utem.edu.my/webopac20/Record/0000106632

Quit rent is a major income for states in Malaysia. Hence, quit rent revenue projection is a crucial component for yearly state budget presentation to ensure the sustainable physical and development in the particular state, as well as throughout the whole country. Identifying predictors is essential to accurately predict the revenue for the next coming year. Current practice of quit rent revenue projection in the state of Negeri Sembilan, Malaysia is to increase the past year revenue by a certain percentage according to the performance indicators in the state. This manual prediction has posted an overrated projection every year. Hence, a more intelligent quit rent revenue estimation technique is needed to automate and improve the projection. This project aims to analyse three benchmarking techniques, namely the Neural Networks (NN), Support Vector Machine (SVM) and Logistic Regression (LR) techniques in quit rent revenue estimation. The studies follows the data science methodology using experimental approach starting from problem formulation to data preparation, model building, results analysis, and concluding on the research findings. The experiment data was built on the quit rent payment transaction in the year of 2015 from the state of Negeri Sembilan, Malaysia. The indicators of account active duration, category of land use, arrears, and late payment charges were used as conditional features. The learning models were built to first predict the payment status before estimating the total quit rent revenue for the year of 2016. The estimation results were compared with actual results and further analysis in details using the performance measures of classification accuracy, precision, weighted mean precision, recall, weighted mean recall, and root mean square error (RMSE). The analysis showed that all the three estimation models have demonstrated good performance. The LR model has achieved the best payment status accuracy of 94.78% followed by the NN model at 94.72% and the SVM model at 91.57%. However, the measurement of RMSE has showed a slight different. The NN model has the closest estimation to the actual total amount of quit rent revenue in the coming year with only 2.07% estimation error in Ringgit Malaysia, while the LR and SVM models were recorded with 2.10% and 4.69% difference respectively. In summary, both the LR and NN models are good to be used as the quit rent revenue estimators. However, all the three models are not able to predict the minority class of payment done after yearly quit rent estimation in October because of imbalance data problem. Further research should focus on treating the imbalance data problem before feeding the data into the learning model. Besides, improving the prediction strategy by taking into account the payment trend and behaviors of land owners is another research direction worth to follow.