Massive stars, although rare, are of key importance for understanding the evolution of the Universe and are the progenitors of supernovae, gamma-ray bursts and even gravitational-wave events. An important parameter of massive stars is their mass loss rate, which has a significant impact on their evolution. Despite its importance, mass loss from massive stars and especially episodic mass loss, is poorly understood. Spectroscopy is an important tool to derive stellar properties, but is a time-consuming process. Instead, the availability of large datasets from photometric surveys, provide data for thousands of sources. Hence, a machine-learning tool has been developed, within the ASSESS project (http://assess.astro.noa.gr/), to automate the classification process of evolved massive stars. Using the machine-learning classifier, which is based on mid-IR (Spitzer) and optical (Pan-STARRS) photometry, we have obtained spectral classification of about 1.1 million sources in 25 nearby galaxies. Our study focuses on massive stars in the Andromeda galaxy, for which we have predictions for ~800K sources. We used the Gaia Andromeda Photometric Survey (GAPS), which provides a significant coverage of the galaxy and offers spectral types and photometric time series. By performing a cross-match between the sources from GAPS (~1.2M sources) and our Spitzer catalogue (~800K sources) we identified approximately 26K common sources. For those we calculated the median absolute deviation to analyze the variability, to identify new variables and constrain the spectral class predictions. We plan to verify our results about the variable stars using the published data products from Gaia DR3. In total the GAPS will help us validate the performance of our machine-learning classifier and gain a deeper understanding of the nature of these objects when variability is considered. Our findings could also be used to refine the machine-learning model and improve its accuracy in classifying massive stars.