A novel comparative investigation of the effect of the number of neurons on the predictive performance of the artificial neural network: An experimental study on the thermal conductivity of ZrO2 nanofluid

In this study, the effect of the number of neurons on the predictive performance of artificial neural networks (ANN) has been investigated using experimental data. For this purpose, 6 different ANN have been developed by using a total of 60 experimental data of ZrO2/water nanofluid obtained from the literature. In ANN developed with the number of 5, 10, 15, 20, 25, and 30 neurons, all other parameters have been kept constant, and the effect of only the number of neurons on the prediction performance has been investigated. The performance of each ANN has been calculated separately and then their performance has been analyzed by comparing them with each other. As a consequence of the study, it has been seen that the model with the most ideal predictive performance has been developed with 5 neurons with an average error rate of 0.001%, and the highest margin of error the model has been developed with 15 neurons and had an error rate of 0.026%. In the light of the obtained data, it has been concluded that ANN are generally high performance predictive tools, and it is not possible to reach a standard correlation to regulate the number of neurons to be used in the optimization of ANN.

Automated summary

This study investigated the impact of the number of neurons on the predictive performance of artificial neural networks using experimental data. The results showed that increasing the number of neurons could improve performance, but adding too many neurons does not necessarily enhance predictive capabilities.