A data-driven framework for failure probability prediction of water supply pipelines: Integrating feature selection and homogeneous grouping