Point and interval prediction of aircraft engine maintenance cost by bootstrapped SVR and improved RFE

Maintenance cost of aircraft engine is an important component of aircraft operation cost. The prediction of aircraft engine maintenance cost can provide decision support for airline to make reasonable maintenance plan and maintenance fund management. Considering that the prediction accuracy of engine maintenance cost is not high in the case of small samples, this paper proposes a bootstrapped support vector regression (SVR) prediction method based on improved recursive feature elimination, which realizes the point and interval prediction of engine maintenance cost in aircraft operation phase. First, the recursive feature elimination (RFE) is improved and then combined with SVR to select feature subsets. Second, particle swarm optimization (PSO) algorithm is applied to optimize the improved RFE-SVR model (IRFE-SVR) parameters. Finally, the point and interval estimates are obtained by bootstrapped IRFE-SVR. To demonstrate the performance of the bootstrapped IRFE-SVR, experiments on UCI and a real case study of engine maintenance cost prediction are conducted. The results on UCI and real datasets show that the bootstrapped IRFE-SVR method has high accuracy and reliability.