Indonesia scholarship selection model using a combination of back- propagation neural network and fuzzy inference system approaches

Abstract

"Bidik Misi" (BM) is a famous scholarship offered by Indonesia Government which has two main qualification aspects: economic needs and academic performance. Since BM scholarship was initiated in 2010, the number of candidates has increased year by year despite limited quota, causing difficulties in selecting the most qualified candidates. This research is proposing a new selection model by combining Back-Propagation Neural Network (BPNN) as candidate's predictor and Fuzzy Inference System (FIS) as candidate's selector. BPNN is used to classify candidates into two, and three recommendation classes then perform FIS on priority set of candidates which is derived from the intersection of classification's result of two and three recommendation classes to make candidate's ranking. Closed test results show that for assuming quota of 50% from total 625 candidates or quota = 312, system can choose 267 highly recommended and reject 268 non-recommended candidates while 45 are identified as moderate recommended, and 45 other are identified as candidates of lower priority with selection accuracy of 85.6% better than five previous similar works. The open test results provide accuracy average of 86.88% and the lowest accuracy of 84% which is close enough to the closed test's accuracy of 85.6% with the difference around 1%, and a small standard deviation value of 2.37 indicates stability of system performance in the selection process. The lowest and highest accuracies of the open test results indicate the proposed system can predict and select BM scholarship's candidates in the future with satisfactory accuracies from 84% up to 90%; thus the proposed system can be trusted as a proper model for BM scholarship selection process.