Proactive product warranty service planning and control: unravelling the boons of customer-generated content and multi-frequency analyses

Abstract

Proactive warranty service problems planning and control serve a crucial role in service continuity and sustainability. In this regard, for the majority of manufacturers, predicting warranty claims for complicated goods is a reliability problem. The prediction of warranty claims is a complex task that also inherently possesses ambiguity, uncertainty, and volatility due to various factors. This study tries to overcome these challenges by providing a unique hybrid forecasting technique and experimenting with the use of social media data to improve the accuracy of warranty claims forecasts. We have established a collaborative partnership with a provider of automobile product services to curate a comprehensive dataset comprising two distinct types of data: (1) internal operational data and (2) social media data sourced from Twitter and Facebook. The proposed algorithm exhibits a high level of proficiency in predicting warranty claims and demonstrates robustness in effectively handling various forms of time series data that display multifarious statistical characteristics. Empirical Wavelet Transform (EWT), Long Short-Term Memory (LSTM), Convolutional Neural Network (CNN), and a Bayesian optimisation algorithm for optimising the shaped Deep LSTM-CNN hyperparameters make up the proposed approach. It is shown that using EWT to provide multi-frequency analysis, LSTM-CNN combination to capture both short- and long-term memory of the data, and the utilisation of social media data significantly improve the efficacy of out-of-sample predictions and counteract the impact of the “warranty data maturation” phenomenon. The suggested algorithm exhibits a decrease in Root Mean Square Error (RMSE) ranging from 29.13% to 90.64% compared to other advanced benchmark models.