• Adi Sucipto


One of the most crucial problems which are faced by financial institution is high level of non-performing loan.
Many analysts have attempted to solve this problem by various strategy and method, but this problem still
persists until now on. With regard to the spreading of credit cooperation, now it is time for the credit
cooperation manager to minimize the level of non-performing loan to under 5 % as the Indonesia bank stated.
On the other hand numerous credit cooperations still cannot hire an analyst because it is expensive, while the
decision to approve or not a credit is a key point which determines the probability of non-performing loan.
Some requirements which relevant to the credit approval are: amount of loan, time, earning, number of family
and personal credit track record. This research uses the data from the “Artha Abadi” Jepara credit cooperation
from the year of 2010 to 2014, the number of credit data as many as 2010 which consist of performing and nonperforming
loan. This research use the data mining classification method with the model of neural network
which is deployed to assess the data processing accuracy with the rapid miner and then continued with the
measurement use the confusion matrix, ROC curve. The result of the neural network algorithm after the test of
confusion matrix, ROC curve and T-test show a very high accuracies value, and the dominant value of AUC and
algorithm. The value of accuracy is 96.95 % with the number of AUC equal to 0.981 and the result of the T test
show that the neural network algorithm is the most dominant result. Finally, this research reveals that the high
level accuracy of the credit analysis can be done with the relatively cheap cost of computation.
Key words: Neural Network, Accuracy, Confusion Matrix, ROC Curve and T-Test

DB Error: Table './ojs/metrics' is marked as crashed and last (automatic?) repair failed