Utilization of Machine Learning to Predict the Surface Tension of Metals and Alloys

As technology progresses, predictive solutions created by computer generated algorithms are becoming more and more viable. The purpose of this study is to test the predictive capabilities and their values of three different types of predictive algorithms, a multi-variable linear regression algorithm, a nonlinear random forest model, and a TensorFlow deep learning neural network model. To compare each algorithm, we used the surface tensions of the molten pure metals, copper, bismuth, and silver, as well as the copper-bismuth, and copper silver molten alloys. The surface tensions were then compiled into data sets meant for training and testing the algorithms predictive capabilities. Throughout this study, we considered how each algorithm could be corrected in ways to increase its predictability without over-constraining the algorithm to satisfy only these data sets. At the end, it became apparent that although the predictions of each algorithm were able to get to a fairly decent accuracy, the random forest model proved to be the best and most useful algorithm for surface tensions.