Predictions of apple bruise volume by using RBF artificial Neural network and comparison it with regression

Bruising damage caused by impact form the main part of all kinds of mechanical damages after harvesting fruits. The prediction of bruise volume is essential for the applicability of the discrete element method to simulate bruise damage during fruit transport and handling. Bruise prediction models can provide useful information about the influence of fruit factors (e.g. ripeness) on bruise susceptibility, leading to recommendations for fruit handling. Bruise prediction models were constructed for the Golden Delicious apple. Bruise volume was used as a measure for apple bruising. The purpose of this research was to evaluate RBF artificial neural network capability in predicting apple bruise volume. The study was conduct using empirical data on 120 apples. Optimal parameters for the RBF neural network were select via a trial and error procedure on the available data. In order to evaluate the performance of the RBF model in the prediction of apple bruise volume, some statistical tests, such as comparisions of the means, variance, statistical distribution as well as linear regression were used between the actual data and the prediction data using RBF neural network. Results showed that in training phase and test phase P-value was greater than 0.9, indicating that there was no significant difference between statsitcal parameters such as average, variance, statistical distribution. This results suggest that RBF neural network can learn bruise volume model very well. Results showed that the predicted values and actual values of apple bruise volume fitted very well (R2 > 0.9), with mean absolute percentage error (MAPE) less than 2.82%. There was no significant difference between the actual values obtained from apple bruise volume and predicted value calculated by models. Also the results reaveled that the model established by RBF was more accurate than regression model to predict apple bruise volume.