Classification of fermentation performance by multivariate analysis based on mean hypothesis testing

Area of Science:

• Biotechnology and Bioprocess Engineering
• Data Science and Machine Learning in Industrial Processes

Background:

• Multivariate analysis is widely used for feature capturing, physiological state recognition, fault detection, and bioprocess control.
• A significant challenge is identifying the most informative process variables for multivariate analysis a priori.

Purpose of the Study:

• To develop an effective method for selecting informative process variables from historical data for multivariate analysis.
• To enhance the performance of multivariate analysis models by optimizing input variable selection.

Main Methods:

• Employed Mean Hypothesis Testing (MHT) as a premodeling step for variable selection.
• Classified fermentation datasets into 'good' and 'bad' classes based on MHT results.
• Applied selected variables to Principal Component Analysis (PCA) and Artificial Autoassociative Neural Network (AANN) models.

Main Results:

• MHT effectively identified discriminating process variables from historical fermentation data.
• The selection of variables using MHT significantly enhanced the overall effectiveness of subsequent multivariate analyses.
• Improved performance was observed in PCA and AANN model creation when using MHT-selected variables.

Conclusions:

• Mean Hypothesis Testing is a valuable premodeling technique for selecting optimal input variables in multivariate analysis.
• This approach improves the accuracy and reliability of physiological state recognition and bioprocess performance analysis.
• The MHT method offers a robust strategy for enhancing bioprocess monitoring and control.