Machine learning models for prediction of droplet collision outcomes

Predicting the outcome of liquid droplet collisions is an extensively studied phenomenon but the current physics based models for predicting the outcomes are poor (accuracy ≈ 43%). The key weakness of these models is their limited complexity. They only account for 3 features while there are many more relevant features that go unaccounted for. This limitation of traditional models can be easily overcome through machine learning modeling of the problem. In an ML setting this problem directly translates to a classification problem with 4 classes. Here we compile a large labelled dataset and tune different ML classifiers over this dataset. We evaluate the accuracy and robustness of the classifiers. ML classifiers, with accuracies over 90%, significantly outperform the physics based models. Another key question we try to answer in this paper is whether existing knowledge of the physics based models can be exploited to boost the accuracy of the ML classifiers. We find that while this knowledge improves the accuracy marginally for small datasets, it does not improve accuracy with if larger datasets are used for training the models.