Data mining-based algorithm for pre-processing biopharmaceutical manufacturing records

• Published in: Computer Aided Chemical Engineering Vol. 44; pp. 2263 - 2268
• Main Authors: Casola, Gioele, Siegmund, Christian, Mattern, Markus, Sugiyama, Hirokazu
• Format: Book Chapter
• Language: English
• Published: 2018
• Subjects: GMP; Natural language recognition; Supervised machine learning; Unsupervised machine learning; Natural language recognition; Unsupervised machine learning; Supervised machine learning; GMP
• ISBN: 9780444642417; 0444642412
• ISSN: 1570-7946
• DOI: 10.1016/B978-0-444-64241-7.50372-4

In this work, a data mining-based algorithm is presented for the pre-processing—e.g., noise removal, batch isolation—of continuously measured historical records of biopharmaceutical manufacturing. The algorithm applies approximate string match and decision tree classifiers to remove noise from commercial production data automatically. Single batches are isolated using k-means clustering, after which algebraic semantic is used to characterize whether the data points within a batch describe the normal process execution or failures. The algorithm was applied to a dataset containing two months of manufacturing data in a cleaning and sterilization process. The performance of the algorithm was evaluated, resulting in a yield of 95 %, a mean time deviation of 1.3±4.6 %, and a rate of misclassification of 1.5 %, which showed a high performance of the algorithm. This study supports the introduction of data-driven automation approaches as well as smart manufacturing in pharmaceutical manufacturing.