Assessing the Auto Associative Network Approach for Prediction in Civil Engineering Databases

Auto-associative networks are a type of Artificial Neural Network (ANN) architectures that has been used in a variety of engineering areas for the past two decades. In auto-associative networks, the knowledge to be extracted from a database is the identity function. In other words, this particular network is trained to reproduce its inputs and output(s). Due to the fact that the network is optimized on inputs, as well as outputs, obtaining highly accurate results can be challenging. In this study, auto-associative network was explored using seven civil engineering databases from various applications and with a range of data types. The architecture of the auto-associative networks was developed with only three layers - input, hidden, and output layers - in order to maintain the generalization capabilities. Only the output was considered when assessing the statistical accuracy measures. A traditional ANN model was developed for each database to provide an initial estimate of the output. Then these estimates and the inputs were used to develop the auto-associative network. The auto-associative network improved the statistical accuracy measures for some databases relative to the traditional ANN approach. Overall, the auto-associative network yielded promising results and can be applicable to civil engineering databases.