Willingness-to-pay Prediction Based on Empirical Mode Decomposition

Long-term prediction of customer preferences is becoming essential for effective product portfolio design in broad industrial sectors such as automotive, aerospace, consumer electronics, where typical concept-to-release times are long (24-60 months). However, nonlinear and nonstationary evolutions of customer preferences hinder accurate prediction of the futures of customer preferences. This paper presents a two-step prediction approach based on Empirical Mode Decomposition (EMD) to forecast customer preferences over extended time-horizons. The advantage of EMD is that this method can be used to decompose a nonstationary time series into a finite number of components called intrinsic mode function (IMF). This property helps in isolation of trend and noise components (detrending and denoising) from a nonstationary process. However the presence of edge artifacts limits the use of EMD for prediction applications. A key aspect of our approach is that it takes advantage of the linear phase property of Hilbert-Huang Transform (HHT) to address this artifact, thus extend EMD for long-term prediction applications. The empirical results suggest that EMD based prediction can significantly improve prediction accuracy in terms of RMSE (36%) and R 2 (30%) for long-term prediction, compared to classical and advanced time series techniques.