Designing Water-in-Oil Emulsion Using Microfluidic Systems Through Machine Learning

Due to the instability of water-in-oil (W/O) emulsions, there are a limited number of studies dealing with this research topic. Generating size-tunable and stable emulsions is a challenging yet crucial task because of the wide range of applications that emulsions encompass. This study proposes a new machine learning (ML) model to predict droplet sizes of water-in-mineral oil emulsions generated by quartz microfluidic chips under various production conditions. The microfluidic system generated highly monodisperse emulsions with defined sizes ranging from 5 to 20 µm. The proposed ML model is based on a stacking ensemble that integrates predictions from several ML-based models, including random forest regression, gradient boosting, support vector regression, XGBoost, and artificial neural network into a single meta-learner model to accelerate the prediction and improve the model's generalizability. We applied different data augmentation techniques to increase the dataset's size and diversity. This involves creating new data instances by applying transformations to existing data points and duplicating existing data with perturbations. The effectiveness of these models is evaluated on the basis of the real dataset and the synthetic data. The obtained results demonstrate that the proposed ML model is based on a stacking ensemble is highly effective compared with other ML models. In addition, the proposed model can reduce the need for microfluidic experimentation and facilitate the adoption of microfluidics in life sciences.