"MiRiGS v1.0: Simulated Microring Gas Sensing Dataset"

"MiRiGS v1.0_Simulated Microring Gas Sensing Dataset is a simulated microring resonator gas sensing dataset designed for benchmarking feature selection and lightweight classification in consumer IoT. It emulates a silicon microring array with polymer coated rings selective to ammonia (NH3), methane (CH4), and carbon dioxide (CO2). The release contains 1,200 instances across eight exposure classes and 30 candidate features that include core physics and engineered interactions, plus a concise data dictionary and a validation report. Core variables cover optical wavelength, normalized transmittance, ring radius, effective index, resonance shift, and gas concentrations; engineered terms (for example squared features, ratios, and cross products like wavelength times NH3 concentration) capture nonlinearity and cross effects that arise in photonic sensing. Signals are generated from standard microring behavior, where resonance depends on effective index and path length, and gas\u2013polymer interaction follows a simple Henry or Langmuir type response. Notch depth varies with a quality factor, and measurement noise combines white and low frequency components to reflect typical optical spectrum analyzer behavior. Parameter draws and class labels are reproducible via fixed seeds. Typical ranges are: wavelength 1400\u20131600 nm, transmittance 0\u20131, ring radius 5, 10, or 15 um, effective index 2.45\u20132.55, resonance shift up to about 1.2 nm, and gas concentrations from sub-ppm to a few thousand ppm. The validation report provides per column min, max, mean, standard deviation, and pass or fail checks against declared bounds; the table has no missing values or duplicate rows. Intended uses include evaluating wrapper and hybrid metaheuristics under accuracy versus subset size trade offs, studying robustness with class imbalance and mild spectral drift, and supporting teaching and reproducible experiments in photonic sensing and edge AI. As a simulation informed by established models and plausible ranges, the dataset complements, but does not replace, device specific calibration."