The　deep　forest　(DF)　model　is　built　using　a　multilayer　ensemble　of　forest　units　through　decision　tree　aggregation.　DF　presents　characteristics　of　an　easy-to-understand　structure,　is　suitable　for　small　sample　data,　and　has　become　an　important　research　direction　in　the　field　of　deep　learning.　These　attributes　are　particularly　suitable　for　the　modeling　of　difficult-to-measure　parameters　in　actual　industrial　process.　However,　existing　methods　have　mainly　focused　on　the　problem　of　DF　classification　(DFC)　and　cannot　be　directly　applied　to　regression　modeling.　To　overcome　these　issues,　a　survey　on　the　DFC　algorithm　is　presented　in　terms　of　constructing　a　small　sample　data-oriented　DF　regression　(DFR)　model　for　industrial　processes.　Hence,　a　survey　on　the　DFC　algorithm　is　presented　to　construct　a　small　sample　size　of　the　data-oriented　DF　regression　(DFR)　model　for　industrial　processes.　First,　principle　and　properties　of　DFC　are　introduced　in　detail　to　demonstrate　the　non-neural　network　deep　learning　model.　Second,　methods　of　DFC　are　discussed　in　terms　of　feature　engineering,　representation　learning,　learner　selection,　weighting　strategy,　and　hierarchical　structure.　Furthermore,　related　studies　on　decision　tree　algorithm　are　reviewed　and　future　investigations　on　DFR　and　its　relationship　with　deep　learning　are　discussed　and　analyzed　in　detail.　Finally,　conclusions　and　the　future　direction　of　this　study　for　industrial　process　modeling　are　presented.　Developing　a　DFR　algorithm　with　characteristics　of　dynamic　adaptive　and　interpretation　abilities　and　lightweight　structure　on　the　basis　of　actual　industrial　domain　knowledge　will　be　the　focus　of　our　follow-up　investigation.　Moreover,　the　existing　research　results　of　DFC　and　deep　learning　can　provide　guidance　for　the　future　investigations　on　the　DFR　model.