5. Concluding Remarks
Auto-associative network approach with backpropagation learning algorithm was explored by using seven civil engineering databases.
Effects of input parameters on the output based on the statistical evaluation criteria was utilized to determine the optimal architecture
of the neural network models, while mapping input parameters on the output layer as well. The auto-associative network method utilized
the outputs from static ANN models along with the input parameters to generate new improved results, as well as to provide reflection
of the predicted outputs and input parameters. Due to the fact that auto-associative network is optimized on inputs and output(s), the
statistical accuracy measures of the outputs were not expected to be as reliable as ANN models. However, the results indicated that for
few cases auto-associative network can perform better. The auto-associative network did not perform well on most of the databases in
terms of error reduction, but discovered the relationship between inputs and output. Even though the results from auto-associative
network are not comparable with those obtained via other approaches, they are still considerably promising. It is noteworthy to mention
that auto-associative network can not only be utilized to generate outputs, but can also be used for verification of the missing values in
input parameters.