A Computational Physics–Based Machine Learning Modelling of Multiphase Flow Dynamics for Crude Oil Percentage Prediction Using Water Cut and Sediment Indicators

Existing crude oil percentage prediction methods often rely on direct measurements and historical data, neglecting the coupled multiphase characteristics of oil–water–sediment systems, which limits predictive accuracy. This study develops a computational physics–based machine learning model integrating key multiphase production parameters, including water cut, basic sediment, and BS&W, using samples from PT. Pertamina Puspa Field Jambi. Data were split into two sets: one for model development and one for validation to prevent overfitting. Linear Regression, Support Vector Machine (SVM), and Random Forest algorithms were applied, with Linear Regression achieving the best performance. For the test dataset, the model yielded a Mean Absolute Error of 0.022168, a Mean Squared Error of 0.001227, and an accuracy of 0.99877, demonstrating precise capture of multiphase interactions. The proposed computational physics–based modelling framework provided improved predictive reliability and consistency. Correlation analyses indicated a coefficient of determination (R²) of 0.99 and a perfect negative correlation (r = −1) between BS&W and oil content, showing that higher BS&W corresponds to lower oil percentage. This framework offers improved predictive reliability and consistency for crude oil quality assessment.