Development of Stepwise Forecasting Experimental Design Methods Based on AI Technologies

Purpose: The objective of this paper is to develop forecasting experiment procedures increasing the efficiency and effectiveness of experiments by combining DoE (Design of Experiments) and AI (Artificial Intelligence) algorithms to reduce unnecessary cost and period in phase of animal experiments in the field of new drug development.Methods: A methodology utilizing AI algorithms like k-NN and XGBoost for interpolating outliers and missing values of DoE results and for predicting results at remaining experimental points of FD (Factorial Design) based on FFD (Fractional Factorial Design) results is proposed in a stepwise experimental design methods.Results: In this case study, a proposed methodology utilizing AI algorithms for predicting results at remaining experimental points show performance of XGBoost is better than k-NN and the predicting results are significant. Especially, when predicting results at remaining experimental points of FD (Factorial Design) based on FFD (Fractional Factorial Design) results, predicting results are sensitive from whether or not data of center points. This proposed methodology can reduce the cost and period for retesting by utilizing an appropriate AI algorithm in a stepwise experimental design methods.Conclusion: Combining DoE based on traditional statistical methods with AI algorithms for predicting experimental results is shown that a stepwise experimental design methods can become more efficient and effective.