Neuro-Fuzzy Predictions of Construction Site Completion Dates

The results from two types of multi-layer perceptron artificial neural networks (Matlab R2015a was used) were compared. The first one, with only one neuron in an output layer having the value of delay in completion date of building site. The output layer of the second artificial neural network is created by three neurons. These three values represent the same delay, but in a form of three values of membership functions to fuzzy sets. In order to evaluate the accuracy of predictions, the mean squared error was used. It was necessary to find the best method of defuzzyfication predicted delays to compare the results from these two, aforementioned artificial neural networks. The level of prediction accuracy measured by mean squared error was discussed, too.