<p>Pipeline transportation, an important aspect of petroleum exploitation and processing, is often prone to flow assurance problems such as leakages and blockages. Most existing analytical and empirical models for predicting leakage of liquid in petroleum pipelines are either inaccurate or complex for routine multiphase field operations. The importance of using computational techniques for analysing and modelling flows in pipelines has long been recognised. This study was, therefore, designed to develop accurate and data-driven models for predicting leakages during multiphase flow. A horizontal flow loop was designed and constructed using a 38-meter-long, 25.4-mm inner diameter PVC pipe rated for 450 psi. It simulated multiphase flows of kerosene (L₁), diesel (L₂), water (L₃), and sand (S). The setup included upstream and downstream valves, mixing and separation tanks, and calibrated instruments for pressure, flow rate, and temperature measurement. Leakages were simulated using orifices of varying diameters along the pipeline. The rig was evaluated using the Colebrook-White equation. The flow was modelled using machine learning techniques, including Linear Regression, Decision Trees, Random Forests, and Gradient Boosting. The Random Forest Regressor excelled at predicting multiphase flow parameters (R² = 0.9986). Gradient Boosting performed best for leakage prediction (R² = 0.9789). Classifier models accurately determined No-leakage, Leakage and number of leakages (F1 = 1). The study demonstrated that Gradient Boosting and Random Forests could be reliable real-time tools for leakage detection and flow monitoring in multiphase flow. Their scalability and independence from theoretical assumptions make them suitable for industrial applications, particularly in environments with multiphase flows.</p>