Integrating machine learning into a crowdsourced model for earthquake-induced damage assessment