Structural optimization of neural network for data prediction using dimensional compression and tabu search

Abstract

In the traditional procedures, data classification with a high degree of accuracy by neural networks requires heuristic structural optimization by using expert knowledge. However, the optimization procedure takes an immense amount of time and effort. Additionally, high-dimensional data is difficult to classify for many analysts, thus, it would appears that accuracy of data classification grows higher by proper selection and dimensional compression of input data. This study suggests new procedure for data classification by using neural networks. For dimensional compression of input data, the suggested procedure uses sandglass type neural networks, and tabu search algorithms are applied for input data selection and structural optimization of union between a sandglass type and a feedforward neural networks.