This paper proposes a machine learning based prediction model construction methodology for the virtual metrology in LCD manufacturing processes. The proposed prediction model construction methodology consists of four major steps; 1) data preprocessing, 2) feature selection, 3) deep learning model design, and 4) model validation. To extraction effective predictor variables at the feature selection stage, this paper employs three techniques including relative weight method, random forest method, and genetic algorithm. The constructed prediction model has been applied to LCD manufacturing data, and shows reasonably acceptable prediction accuracy which is higher than 90%.
