Statistical and machine learning applied to the universal consolidated database to predict heat transfer coefficient in flow condensation

Due to complex physical mechanisms, and a general lack of accurate modeling techniques, two phase convective heat transfer is hard to characterize. In this article we show the general effectiveness of statistical and machine learning techniques for predicting two-phase heat transfer coefficients. This is a comprehensive study that: performs a full exploratory data analysis (EDA) to determine the range by which we may apply these models, derives an optimal data set as a basis for predicting heat transfer coefficient, tests all models on an independent data set, and assesses model performance on multiple metrics. We find that random forest methods, gradient boosted machines, and multi-layer perceptrons (deep neural networks) perform the best in predicting heat transfer coefficient on the given data set. We find that in general, the increased data base predicts better due to the inclusion of a few important thermophysical and flow properties.