Title: Machine learning and uncertainty analysis for remaining value estimation Authors:  Ieva Dunduliene - Kaunas University of Technology (Lithuania) [presenting]
Robertas Alzbutas - Kaunas University of Technology, LEI (Lithuania)
Abstract: The estimation of the remaining value is gaining more attention, especially in the context of sustainability, engineering, and industry. Furthermore, the remaining value estimations present additional challenges related to uncertainties, subjective information integration, and exogenous feature dependencies. This task is nontrivial and complicated by the absence of a gold standard for estimating and comparing the calculated remaining values. The challenges especially arise while trying to estimate or/and validate the remaining value of the refurbished products. To address these challenges, it is proposed to integrate machine learning (ML) methods and uncertainty analysis to ensure the accuracy and risk minimization of the remaining value estimation. Machine learning methods are widely utilized in the remaining value estimation process due to the methods' ability to detect and identify relationships and hidden patterns among variables. Through case studies, the effectiveness of the presented approach is proven by providing interval estimates of the remaining value calculations that incorporate uncertainty assessment. This allows decision-makers and consumers to make well-informed decisions when considering the purchase of refurbished products. In conclusion, a framework is presented that integrates ML and uncertainty analysis for remaining value estimation by combining multiple ML models into one and facilitating the quantification of uncertainty in the results.