Comparison of BB and BDLRF Algorithms in the Training MLP Neural Network for Tractor Repair and Maintenance Costs Prediction

The farm manager must be able to predict repair and maintenance costs. These costs can have significant impacts on replacement decisions. The purpose of this research was to evaluate MLP neural network capability in predicting tractor repair and maintenance costs in terms of with regatta tractor age. The study was conducted using field data on 17 Messy Ferguson tractors from Astan Ghodse Razavi Agro-Industry. In order to train the network, collected data were preprocessed. MLP network trained by both BB and BDLRF learning algorithm. The optimal parameters of the network were determined by a trial and error procedure. Statistical comparisons of actual and predicted test data in the test phase were carried out. The results revealed that the neural network model was capable of predicting the repair and maintenance costs with a mean absolute percentage error of less than 3.37. Also, it was found that BDLRF algorithm in comparison to BB algorithm had a better performance.