Cost estimation for general aviation aircrafts using regression models and variable importance in projection analysis

Abstract

Accurately estimating the development cost of general aviation aircraft plays a key role in devising the best strategy for corporates. However, studies on cost estimation for general aviation aircraft are limited. Development cost data of general aviation aircraft are commonly multi-collinear and the sample size is small, which may cause the estimation performance of the developed model to be poor. To address this issue, a combination of the variable importance in projection (VIP) analysis method and regression models is proposed. The VIP analysis method, applied to strong correlation data and small sample size, is utilized to select the most influential independent variables. The combined regression models tested using the selected variables include a partial least-squares regression (PLS) model and a back-propagation neural network (BPN) model. The PLS and BPN model are established respectively by utilizing the unfiltered raw variables for comparison. To verify the accuracy and feasibility of the proposed method, a case study utilizing general aviation aircraft cost data is presented. The results suggest that the VIP analysis method could be utilized for variable selection, and the proposed PLS model combined with VIP analysis has better prediction accuracy than that of the pure regression models and the combined BPN model, with MSE, MMRE, and R2 values of 1.07, 4.1% and 98.32%, respectively. Therefore, the VIP analysis method combined with the regression model can be effectively used to estimate the preliminary cost of general aviation aircraft. More importantly, this work provides a feasible method to improve the accuracy of cost estimation for general aviation aircraft.