Multi-target regression via target combinations using principal component analysis

Data-driven methods have become increasingly widespread in the chemical industry; however, these methods require sufficient data for effective implementation. Moreover, collecting sufficient data can be challenging for new processes, products, batch processes, and other scenarios. To address this limitation, we propose a novel method called the principal component analysis target combination method, combining the Random Linear Target Combinations (RLC) and Principal Component Analysis. The proposed method utilizes the relationships of multiple inspections obtained in a single production cycle to improve the prediction accuracy using the Multi-Target Regression (MTR) approach. Although collecting multiple samples can be time-consuming and expensive, conducting several inspections on a single sample is relatively easy. In chemical products, multiple inspections are usually performed, and relationships, such as trade-offs, between these inspections exist. We first evaluated the effectiveness of the proposed method using 12 diverse real-world benchmark datasets frequently used in MTR tasks. The results show that the proposed method improves the average Relative Root Mean Squared Error by 3.16 % compared to single-target regression (Gradient Boosting) and improves by 2.0 % compared to RLC, upon which the proposed method relies. Subsequently, we evaluated the method on a real dataset from our company comprising formulation-design data for developing new elastic sealant products. The proposed method outperformed the other methods, confirming its effectiveness and ability to utilize multiple inspection data in chemical plants.