On the accurate construction of machine learning models to predict speed of sound in liquid siloxane

Abstract

The speed of sound in liquid siloxanes is a critical factor that has an important role in various scientific and industrial applications. This research puts forward artificial intelligence methods of Decision Tree, Adaptive Boosting and Ensemble Learning to create models capable of accurately predicting liquid siloxane speed of sound as a function its boiling point, molar mass, pressure and temperature parameters. An experimental dataset is utilized for this purpose, whose validity is checked via an outlier detection methodology. The findings suggested that nearly all the data is appropriate for the goal of developing a data-driven model. Additionally, it is demonstrated that molar mass if the most influential parameter, negatively affecting liquid siloxane’s speed of sound. The assessment metrics and visual analyses reveal that AdaBoost model is the greatest precise for the estimation job. It attains the greatest R2 value (0.990824) and the least MSE (1,504.315) on the test set, coupled with the lowest AARE% (3.109657). These findings emphasize AdaBoost's exceptional capability to identify intricate patterns and provide accurate predictions, particularly for task of forecasting speed of sound in liquid siloxane. The evaluation study revealed the fact that all the developed models are robust and accurate in predicting the speed of sound. The developed models can be used easily and without experimental tasks which are known to generally be tedious, costly and time-consuming.