TFT-LCD　is　a　kind　of　thin　film　transistor　liquid　crystal　display.　The　sampling　test　method　is　used　to　estimate　the　quality　of　the　whole　sample,　but　this　method　is　not　comprehensive　and　has　no　timeliness.　The　author　hope　that　machine　learning　can　be　used　to　make　a　reasonable　prediction　of　product　quality　through　each　process　data.　This　paper　mainly　adopts　the　combination　of　SVM　and　random　forest　to　form　a　new　method:　random　SVM(R-SVM).　Multiple　SVM　models　were　established　by　random　sampling　and　number　of　features,　and　the　predicted　values　of　multiple　models　were　averaged　to　obtain　the　final　results.　The　evaluation　standard　is　the　mean　square　error(MSE).　Experimental　results　show　that　R-SVM　is　better　than　traditional　machine　learning　algorithms.　As　is　known　to　as　all　the　random　forest　performs　best　in　the　traditional　machine　learning　algorithm.　The　MSE　of　our　experimental　results　is　0.6　percentage　lower　than　that　of　the　random　forest.　The　research　method　of　this　paper　have　brought　new　research　ideas　for　industrial　data　prediction　for　the　future.　It　also　provides　an　opportunity　for　the　combination　of　random　forest　and　other　traditional　algorithms.　©　2019　IEEE.