A smart, low-cost, multigas sensing system based on a mid-infrared (MIR) metasurface microspectrometer and machine learning is presented. The system uses a metasurface filter array integrated with a commercial IR camera, which is compact (≈1 cm³) and lightweight (≈1 g). The machine learning algorithm analyzes data from the microspectrometer to predict the presence of gases. The system detects greenhouse gases CO₂ and CH₄ at concentrations from 10 to 100% with 100% accuracy, and hazardous gases at low concentrations with 98.4% accuracy. Ammonia can be detected at 100 ppm, and methyl-ethyl-ketone (MEK) at its permissible exposure limit (200 ppm). The system is designed to detect multiple gases simultaneously, with the metasurface filter array providing 20 spectral channels across the 6–14 μm wavelength range. The system is tested with four gases: CO₂, CH₄, NH₃, and MEK, diluted with nitrogen. The machine learning classifier (MLC) is trained using data from these gases and achieves high accuracy in identifying them. The system is stable and robust, with a temperature-controlled environment to minimize drift. The MLC uses a support vector machine (SVM) with quadratic kernels for classification. The system demonstrates high accuracy in detecting gases, even at low concentrations, and is suitable for applications such as environmental monitoring, industrial process monitoring, and mining safety. The study highlights the potential of combining machine learning with IR spectroscopy for smart, low-cost multigas sensing.A smart, low-cost, multigas sensing system based on a mid-infrared (MIR) metasurface microspectrometer and machine learning is presented. The system uses a metasurface filter array integrated with a commercial IR camera, which is compact (≈1 cm³) and lightweight (≈1 g). The machine learning algorithm analyzes data from the microspectrometer to predict the presence of gases. The system detects greenhouse gases CO₂ and CH₄ at concentrations from 10 to 100% with 100% accuracy, and hazardous gases at low concentrations with 98.4% accuracy. Ammonia can be detected at 100 ppm, and methyl-ethyl-ketone (MEK) at its permissible exposure limit (200 ppm). The system is designed to detect multiple gases simultaneously, with the metasurface filter array providing 20 spectral channels across the 6–14 μm wavelength range. The system is tested with four gases: CO₂, CH₄, NH₃, and MEK, diluted with nitrogen. The machine learning classifier (MLC) is trained using data from these gases and achieves high accuracy in identifying them. The system is stable and robust, with a temperature-controlled environment to minimize drift. The MLC uses a support vector machine (SVM) with quadratic kernels for classification. The system demonstrates high accuracy in detecting gases, even at low concentrations, and is suitable for applications such as environmental monitoring, industrial process monitoring, and mining safety. The study highlights the potential of combining machine learning with IR spectroscopy for smart, low-cost multigas sensing.