This paper presents a novel non-invasive microwave sensor designed for selective multi-component liquid analysis. The proposed sensor integrates a spoof surface-based whispering-gallery mode resonator coupled to a transmission line, enabling precise detection of volumetric concentrations in liquid mixtures by analyzing spectral responses over a broad frequency range. The study examines five mutually soluble liquids blended in varying proportions while maintaining a constant total volume. A multivariable regression-based machine learning model predicts the volumetric concentration of each component, utilizing resonance frequencies and their corresponding amplitudes as input features. Principal component analysis (PCA) is employed to assess feature significance. The sensor achieves a root mean square error (RMSE) of 0.025 in its predictions. Additionally, an edge computing system incorporating a Raspberry Pi 4 automates real-time data processing, facilitating rapid and efficient liquid composition analysis. The portable setup ensures low-latency making it highly suitable for laboratory environments requiring precise and automated multi-liquid assessment.