Application of Data Mining Techniques to Support Importers and Exporters Delinquency Prediction: The Case of National Bank of Ethiopia
No Thumbnail Available
Date
2009-01
Authors
Journal Title
Journal ISSN
Volume Title
Publisher
Addis Ababa University
Abstract
The application of Data mining is becoming increasingly common in both the private and public
sectors. Industries such as banking, insurance, medicine, and retailing commonly use data mining
to reduce costs, risks, and increase sales.
Foreign currency is a scarce financial resource in Ethiopia. This scarcity calls for the
consolidation and fostering of existing financial management systems to ensure optimum
utilization of the available resource. The National Bank of Ethiopia (NBE) has the task of
monitoring the settlement of importers and exporters foreign exchange commitments as per the
existing directives. Currently in the NBE there are many importers and exporters who are
delinquent and expected to settle their commitment.
Thus, it is the aim of this study to examine the potential applicability of data mining technology
in building a predictive data mining model that helps to predict potentially delinquent or non
delinquent importers and exporters in relation to their utilization of the foreign currency.
To conduct the study, the researcher adopted the Cross-Industry Standard Process for Data
Mining (CRISP–DM) process model. Several predictive classification models were built both in
decision tree and neural network techniques using WEKA software. The best performing model
was chosen by comparing the models using standard evaluation criteria such as accuracy,
precision, recall and interpretability.
According to the evaluation results, both techniques have shown a promising performance.
However, the best models for both export and import transactions were obtained using decision
tree techniques. The decision tree approach brings about 94.02% accuracy in the case of
predicting export transactions and 98.03% for import transactions. Moreover, the models built bythe decision tree show better results in terms of precision, recall and interpretability for both
transactions. Thus, compared to neural network, the decision tree approaches are more applicable
in addressing the research problem.
Accordingly, some important rules are derived using the selected attributes such as,
MethodofPayment, BaseOfShipment, Country_Region, AmtOfBirrIn_Range, Currency,
Validity_period and EconomicSector that are relevant in business decision making.
In general, the results obtained from the study proved the potential applicability of data mining
technology to predict importers and exporters into predefined classes (delinquent and nondelinquent)
based on their transaction characteristics.
Description
Keywords
Data mining is becoming increasingly common