An experimental study on the comparative analysis of the effect of the number of data on the error rates of artificial neural networks

In this study, the effect of the amount of data used in the design of artificial neural networks (ANNs) on the predictive accuracy of ANNs was investigated. Five different ANNs were designed using the experimentally measured specific heat data of the Al2O3/water nanofluid prepared at volumetric concentrations of 0.0125, 0.025, 0.05, 0.1 and 0.2 using the Al(2)O(3)nanoparticle. The developed ANN is a multi-layer perceptron, feedforward and backpropagation model. In each ANN with 15 neurons in the hidden layer, the volumetric concentration (phi) and temperature (T) values were nominated as input layer factors and the specific heat value was estimated as the output value. With the aim of survey the effect of the amount of data on the predicted results of the ANN, a different amount of datasets were used in each developed ANN. In this context, in total 260 data were used in the Model 1 ANN. Subsequently, the total amount of data was reduced by 20% in each developed neural network and 55 data were used in the ANN named Model#5. The results obtained show that ANNs are highly talented of predicting the specific heat values of Al2O3/water nanofluid. However, in the comparisons, it was evaluated that the amount of data used had a share on the prediction performance of the ANN and that the decrease in the amount of data with the prediction performance of the ANN decreased.

Automated summary

The study investigated the impact of the amount of data used in the design of artificial neural networks on the predictive accuracy of ANNs for specific heat values of Al2O3/water nanofluid. The results showed that while ANNs are capable of accurately predicting specific heat values, a decrease in the amount of data led to a decrease in prediction performance.