Time Series Analysis for Fail Spare Part Prediction: Case of ATM Maintenance

Abstract

Prediction of failed spare parts is an interesting issue in inventory management. Our work applied predictive analytic to forecast future amount of failed spare parts. This research used maintenance time series data from year 2013 to 2016 to train and test data for a prediction model. In the preprocessing step, we looked into new features based on historical data set. Then, we added the day of week feature into the example of the data set. The day of week feature had an impact to spare parts prediction model. Moving average and windowing methods were used in the preprocessing phase before sending through the prediction model. Artificial neural network and support vector machine for regression were applied to predict the amount of failed spare parts. The experiments demonstrated the average accuracy of failed spare part prediction. The result represented that the support vector machine for regression showed the best accuracy at 88.24%. SVR yielded the highest prediction accuracy at 92.7%.