The TFT-LCD Quality Prediction Model Based on L1 Feature Selection and SSA-Optimized MLP-XGBoost Ensemble

Accurate TFT-LCD quality prediction is crucial for improving product quality in the semiconductor manufacturing process. To enhance prediction accuracy, this paper proposes a combined prediction model based on L1-norm feature selection and Sparrow Search Algorithm (SSA) optimized Multilayer Perceptron (MLP) and Extreme Gradient Boosting (XGBoost). First, data preprocessing is performed to remove data with a large number of missing values, improving data quality. Then, L1 feature selection is used to filter out features that are less relevant to quality prediction, retaining those that significantly contribute to the prediction. Next, SSA is utilized to optimize the hyperparameters of the MLP and XGBoost models, improving their fitting capabilities. Finally, the MLP and XGBoost models are integrated, and their weights are dynamically adjusted based on their performance during the prediction process, resulting in more accurate predictions. Validation using actual production data shows that the proposed ensemble model reduces the Mean Absolute Error (MAE) to 0.1033, with prediction accuracy improved by 12.49% to 27.97% compared to other models, demonstrating outstanding predictive performance.