framework simultan cluster learn complex data meghana deodhar joydeep ghosh difficult classif regress problem practition segment data homogen group build model group step procedur result simpler interpret action model lossin accuraci problem predict custom behavior product independ variabl natur partit group pivot oper result depend variabl show entri custom product data matrix present model base cluster meta algorithm interleav cluster construct predict model iter improv cluster assign fit model algorithm provabl converg local minimum suitabl cost function framework gener cluster collabor filter model basedco cluster view simultan segment classif regress independ cluster data build model appli wide rang bi modal multimod data easili special address classif regress problem demonstr effect approach problem experiment real synthet data 