Data quality assessment in product failure prediction models

A problem of product failure prediction within the warranty period is presented in a case of a household appliance manufacturer. Predicting product failure within a warranty period is necessary for optimal resources planning. When based on scarce information and intuition, companies reserve non-optimal amount of funds. To address this problem we developed a machine learning model to decrease the prediction error and make the process of warranty claims more transparent. Following the CRISP-DM process, in this paper, we are focusing on data preparation phases. Results show that among 33 attributes, we have identified seven that hold some prediction value, suggesting that the value of the collected data is small compared to its cost. In the future effort should be invested in collecting quality standardised data across markets.