CONSTRUCTING FINANCIAL FRAUD DETECTION MODEL USING MACHINE LEARNING TECHNIQUE: THE CASE OF WEDERA FARMERS MULTIPURPOSE COOPERATIVE UNION

In Ethiopia there are a number of business organizations which are hostile with fraudulent financial activities.This is why in most of business organization t here is lack of automatic system which assists in the detection of such activity. This study tried to predict fraudulent financial activity of wedera farmers' multipurpose cooperative union using data mining techniques and finally build validated prototype for convenient use.For this study out of 7500 financial datasets 5222 financial records were selected to build the predictive model using D e s i g n s c i e n c e r e s e a r c h m e t h o d o l o g y i n t e g r a t e d w i t h d a t a m i n i n g p r o c e s s m o d e l . T h e s t u d y u s e d t w o d a t a m i n i n g a l g o r i t h m s , such as J48 decision tree from tree classifiers and artificial neural network (multilayer perception)from function classifiers and registered best accuracy with J48 algorithm. Since J48 algorithm registered the highest accuracy, it is chosen to integrate weka with java and develop validated prototype system. To do this weka.jar library plays a great role for weka to java connectivity. The financial item "credit of cash" score was found as the most determinant attribute in the union transaction as well as for the financial fraud detection and prevention. In this case, the financial status is fraud if credit of cash is greater than 0.707107 otherwise the rule checks other financial items in order to know whether the financial status is nonfraud or fraud.This result was found since credit of cash serves as the base for a given financial activity. The prototype has achieved 85% system accuracy and 78% accuracy in the case of user acceptance test;but,the major challenge observed in this study is that the natures of the datasets are so difficult to train the model;because,in Accounting and finance rule if a number of attributes exist on a given instance once a time the financial dataset only touch two or three attribute value and the rest left null. So to solve such challenge the study used MATLAB programing language in order to fill null value. Finally even if the prototype scored a capable result, further study is needed by adding additional financial attributes and algorithms in order to enhance and yield a better system performance.