Context. Star formation in the outer Galaxy, namely, outside of the Solar circle, has not been extensively studied in part due to the low CO brightness of the molecular clouds linked with the negative metallicity gradient. Recent infrared surveys provide an overview of dust emission in large sections of the Galaxy, but they suffer from cloud confusion and poor spatial resolution at far-infrared wavelengths.

Aims. We aim to develop a methodology to identify and classify young stellar objects (YSOs) in star-forming regions in the outer Galaxy and use it to resolve a long-standing disparity in terms of the distance and evolutionary status of IRAS 22147+5948.

Methods. We used a support vector machine learning algorithm to complement standard color-color and color-magnitude diagrams in our search for YSOs in the IRAS 22147 region, based on publicly available data from the Spitzer Mapping of the Outer Galaxy survey. The agglomerative hierarchical clustering algorithm was used to identify clusters. Then the physical properties of individual YSOs were calculated. The distances were determined using CO 1-0 from the Five College Radio Astronomy Observatory survey.

Results. We identified 13 Class I and 13 Class II YSO candidates using the color-color diagrams, along with an additional 2 and 21 sources, respectively, using the applied machine learning techniques. The spectral energy distributions of 23 sources were modeled with a star and a passive disk, corresponding to Class II objects. The models of three sources include envelopes that are typical for Class I objects. The objects were grouped into two clusters located at a distance of similar to 2.2 kpc and 5 clusters at similar to 5.6 kpc. The spatial extent of CO, radio continuum, and dust emission confirms the origin of YSOs in two distinct star-forming regions along a similar line of sight.

Conclusions. The outer Galaxy may serve as a unique laboratory for exploring star formation across environments, on the condition that complementary methods and ancillary data are used to properly account for cloud confusion and distance uncertainties.