A neural networks system for predicting Islamic banks performance

Abstract AR

The objective of this research is to use neural networks for predicting Islamic banks performance.

A data sample of twenty six Islamic banks has been collected for the period 1991-1993.

Seven financial ratios were constructed from the data sample.

Kohonen neural network was used first to group the Islamic banks into high and low performance groups using the seven financial ratios for the performance year (1993).

The results of this network have assigned twelve banks to the high performance group and fourteen banks to the low performance group.

Using these results, a three layers Back propagation neural network was built to predict the Islamic banks performance.

The network was trained on the one year prior to the performance year data set while using the performance year data as a test set.

The two year prior to the performance year data set was used as a holdout sample to check the network prediction abilities.

When the network was applied, the results showed high prediction abilities and generalization.

It correctly classified all the Islamic banks on the performance year and on the one year prior data sets.

It also correctly classified 89 % of the banks on the two years prior data set which was unseen by the network.