Dual-Band Microwave Spectroscopy and Machine Learning for Rapid Milk Adulteration Detection

This work presents a novel, low-cost, and portable microwave sensor-based system for the detection and classification of milk adulterants using microwave spectroscopy and machine learning. The sensor operates at two distinct frequencies, 1.25 and 2.4 GHz, and exhibits high sensitivity to dielectric changes caused by adulteration. A set of eight common adulterants, including tap water, mineral water, urea, liquid whey, vegetable protein, soya protein, extraneous water, and sodium carbonate, were tested at concentrations ranging from 5% to 50%. The sensor is operated through a Raspberry Pi, and ADALM-PLUTO Software Defined Radio, and the output is observed on a monitor through a graphical user interface. The transmission response (S₂₁) shows attenuation levels dropping from 0 dB to nearly-18dB as adulterant concentration increases, with clear resonance suppression. The differential resonance shift (Δ f = f₂ - f₁) exhibits a monotonic trend with adulterant percentage, improving sensitivity mapping. Statistical analysis of experimental trials ($N=10$per sample) yielded attenuation mean values with standard deviation error bars, confirming repeatability and robustness. To further strengthen the analytical rigor and decision-making capability of the proposed system, machine learning was employed to extract meaningful patterns from the sensor data and enable reliable classification. A random forest algorithm achieved an overall accuracy of 91.2% and an area under the curve score exceeding 0.7, validated by the confusion matrix and receiver operating characteristic curve. The proposed system demonstrates potential for real-time, noninvasive quality assessment of milk, combining microwave sensing with AI-based decision support to detect and classify adulterants with high sensitivity and specificity.