Application of Feature Selection Method to Error Factor Extraction of Multifunction Peripheral

Multifunction peripheral (MFP) manufacturers provide customers with remote maintenance services, such as supplies provision and automatic firmware updates, to lower customer burdens and to avoid device downtime. Such remote services are required for maintenance so that Japanese machine manufacturers can deliver products to foreign markets, because service bases in overseas locales must cover broader geographical areas than those in Japan. When MFP devices experience a fault, they generally alert users of an error. Although some faults can be solved remotely, there are faults that require an engineer to perform on-site actions. To repair them on-site efficiently, online investigation and pre-assessment of fault factors will be effective. In this paper, we apply the Group Lasso regularization method for logistic regression to select features determined as error factors. We evaluate the engine on two kinds of error examples: those frequently causing alerts in MFP models in the past, and those causing alerts due to part wear. This engine is expected to help engineers determine causal factors of errors.