Multiphase flow Measurement is pivotal in myriad industrial applications, including but not limited to oil and gas production, chemical engineering, and environmental monitoring. Accurate forecasting of multiphase flow dynamics is quintessential for refining process optimization, ensuring safety, and amplifying operational efficiency. This research leveraged a spectrum of non-invasive sensors alongside the drift-flux model to rigorously assess and juxtapose the performance metrics of the two machine learning strategies. The experimental framework was established within a two-storey fluid loop pipeline that managed a biphasic gas-water flow. The integrated sensor array consisted of a dual-plane resistance tomography sensors, electromagnetic flowmeters, temperature sensors, and pressure sensors. Signal from these sensors was transmitted to a data acquisition system called Visualisation and Measurement of Multiphase Flow. The experimental design incorporated 185 unique flow regimes, with water flowrate varying from 1.67×10^-3 to 7.31×10^-3 m³/s and gas flowrate ranging from 8.33×10^-5 to 1.33×10^-3 m³/s. The collected dataset formed the basis for training two novel machine learning models, both featuring advanced feature extractor heads, one utilizing Long Short-Term Memory and the other a Transformer encoder. In the optimal prediction outcome, the relative errors for gas and water flowrate predictions were 2.13% and 0.54%, respectively. The datadriven approach elevates the accuracy of traditional sensors in resolving the complexities of multiphase flow quantification and offers potential solutions for use in real-world industrial applications.