<p>Prediction of the two-phase pressure drop across pipe fittings such as orifices with sufficient accuracy is essential to maintaining pipeline integrity and optimizing operational efficiency. Traditional correlations are often used for pressure drop prediction, yet their accuracy is limited to the examined dataset. This study focuses on predicting the pressure drop in adiabatic air–water two-phase flow through orifices under horizontal and vertical orientations using a data-driven machine learning framework. The dataset consists of nearly 700 flow conditions has been collected from various independent experiments from different works, to represent a wide range of flow conditions of two-phase flow through orifices. The extracted data were then transformed into dimensionless numbers commonly used in correlations of two-phase flow pressure drop across orifices. A dataset preprocessing includes, z-standardization, and then splitting into 0.75 and 0.25 for training and validation has been performed. Several machine learning approaches, namely deep learning, decision trees, random forests, gradient-boosted trees, and support vector machines, were implemented and systematically. In addition, a simplified linear model based on the Lockhart–Martinelli parameter was developed using the same dataset. The results indicate that machine learning models consistently outperform traditional methods in predicting two-phase pressure drops. Among all models, the support vector machines achieved the highest predictive accuracy with an RMSE of 23.349. Feature importance analysis further shows that the use of the Lockhart–Martinelli parameter alone is sufficient to predict the pressure multiplier within a machine learning framework. The proposed simplified linear model also demonstrated better predictive capability compared to traditional correlations with RMSE of 48.83.</p>