Large dataset on Fourier transform near infrared (FT-NIR) spectroscopy of green and roasted specialty coffee: Preprocessed infrared spectra and sensory scores for machine learning-based quality monitoring

A comprehensive dataset on FT-NIR spectra for green and roasted specialty coffee is critically needed to advance predictive models for sensory quality assessment. To address this need, the NIR spectra of green and roasted specialty coffee were obtained using cutting-edge FT-NIR spectroscopy, while sensory quality analysis for roasted coffee beans followed the standardized protocol validated by the Specialty Coffee Association (SCA). This approach was designed to establish a robust dataset for calibrating machine learning-based predictive models of coffee cup quality. FT-NIR spectra were acquired using a Spectrum Two N-FT-NIR Spectrometer equipped with a high-resolution Indium Gallium Arsenide (InGaAs) detector, operating in diffuse reflectance mode. Spectral data were collected over a wavelength range of 12,000 to 4,000 cm^-1 with a spectral resolution of 8 cm^-1, an interval of 4 cm^-1, and 64 accumulated scans per sample. The dataset includes both raw and preprocessed FT-NIR spectra, incorporating baseline correction, Standard Normal Variate (SNV), Multiplicative Scatter Correction (MSC), as well as first and second derivatives to enhance spectral interpretation. One of the main strengths of this dataset is its ability to facilitate non-destructive quality assessment of roasted coffee using both green and roasted FT-NIR spectra. Furthermore, it also enables the prediction of roasted coffee sensory quality based on green coffee spectra, potentially reducing the need for roasting during quality evaluation and streamlining batch screening processes. This dataset enhances research on coffee quality, chemical marker identification, and roasting optimization, supporting both scientific and industrial applications. The dataset is structured into Excel files, systematically organized by processed samples and their replicates, providing a valuable framework for further analysis, model development, and calibration of multivariate statistical models.