Gas-liquid two-phase flow in porous media exists extensively in natural phenomena and numerous industrial processes, and the flow patterns of two-phase flow are crucial for understanding the flow characteristics and modeling the governing mechanisms. However, limited by the porous structure and the transient nature of two-phase flow, the traditional approaches for identifying the flow patterns in porous media often confront great challenge especially in terms of convenience and accuracy. By integrating machine learning techniques, this study proposes an integrated machine learning approach to identify air–water two-phase flow patterns in porous media coupled with the technologies of principal component analysis (PCA) and K-means clustering. Firstly, the time-domain analysis and frequency-domain analysis are carried out for the measured differential pressure signals of two-phase flow in porous media, aiming to extract typical features of two-phase flow in the time-frequency domain. Then, the Principal Component Analysis (PCA) model is developed by taking the typical features of time-frequency domain and the key physical parameters such as gas and liquid Reynolds numbers into account. The first three principal components are selected for dimensionality reduction in the PCA process. Subsequently, classification using K-means clustering enables the identification of both typical flow patterns and key transitional regimes, particularly the bubbly-slug transition. The machine learning approach provides a robust and efficient tool for the rapid identification of gas–liquid two-phase flow patterns in porous media.