A comparison between artificial neural network and random forest on predicting ferrofluids viscosity under magnetic field application

This research study focuses on predicting ferrofluids’ viscosity using machine learning models, artificial neural networks (ANNs), and random forests (RFs) incorporating key parameters; ferrofluid type, concentration of magnetic nanoparticles, temperature, and magnetic field intensity as inputs. A comprehensive database of 333 datasets sourced from various literatures was utilized for training and validating models. The ANN model demonstrated high accuracy, with root mean square error (RMSE) values below 0.033 and mean absolute percentage error (MAPE) not exceeding 3.01%, while the RF model achieved similar accuracy with RMSE under 0.052 and MAPE below 4.82%. Maximum deviations observed were 9.14% for ANN and 16.48% for RF, confirming that both models accurately learned the underlying patterns without overestimating viscosity. Additionally, the ANN model successfully captured intricate physical relationships between input parameters and viscosity when it was used to predict viscosity for random input data, confirming its ability to generalize beyond the training dataset. The RF model, however, showed limitations in extrapolating beyond the range of the training data. This research study demonstrates machine learning models’ effectiveness in capturing intricate relationships governing the viscosity of ferrofluid for different types, paving the way for an improved understanding of ferrofluid’s viscosity behavior.