<p>Machine learning methods (principal component analysis, support vector regression, and partial least squares) along with the traditional ratiometric method were used to calibrate the temperature based on upconversion fluorescence spectra of Ho³⁺ in aluminofluoride glasses in the wavelength ranges 505–560 nm and 605–670 nm for the temperature range from 30 to 105°C. The best calibration models among the methods considered were obtained by partial least squares with searching the combination of moving window partial least squares with the fluorescence spectra of 95MgCaSrBaYAl₂F₁₄–5Ba(PO₃)₂:5%Yb³⁺,0.1%Ho³⁺ glass. The absolute sensitivity of the temperature calibration of 0.021 K^–1 and temperature uncertainty of 0.12 K achieved by machine learning methods significantly exceeded the characteristics of the ratiometric method.</p>