Using an Enhanced Feed-Forward BP Network for Predictive Model Building from Students' Data. Issue 2 (2nd April 2016)

• Record Type: Journal Article
• Authors: Ajiboye, Adeleke Raheem; Abdullah-Arshah, Ruzaini; Qin, Hongwu
• Is Part Of: Intelligent automation & soft computing. Volume 22:Issue 2(2016)
• Page Start: 169
• Page End: 175
• Publication Date: 2016-04-02
• Subjects: Prediction -- data partitioning -- feed-forward network -- back-propagation -- data mining; Artificial intelligence -- Periodicals; Intelligent control systems -- Periodicals; 003.5
• DOI: 10.1080/10798587.2015.1079364
• Languages: English
• ISSNs: 1079-8587

Abstract

Abstract: Feed-forward, Back Propagation (BP) Network is a network structure capable of modeling the class prediction as a nonlinear combination of the inputs. The network has proven its suitability in solving several complex tasks, most especially when trained with appropriate algorithms. This study presents an enhancement of this network with a view to boosting its prediction accuracy. The paper proposed a modification of the data partitioning function in the regular feed-forward network. A predictive model is constructed based on the proposed partition, while the second model is based on the partition of the existing network. Both models are trained and simulated with sets of untrained data. The mean absolute error is computed for both models and their error values are compared. Comparison of their results shows that the enhanced network consistently delivers higher accuracy and generalized better than the existing network in its regular structure; as there was a decrease in error from 0.261 to 0.016. The enhanced network has also shown its suitability in the fittings of models from students' data for prediction purposes.