Galaxy Morphology in CANDELS: Addressing Evolutionary Changes Across $0.2 \leq z \leq 2.4$ with Hybrid Classification Approach

Morphological classification of galaxies becomes increasingly challenging with redshift. We apply a hybrid supervised-unsupervised method to classify $\sim 14,000$ galaxies in the CANDELS fields at $0.2 \leq z \leq 2.4$ into spheroid, disk, and irregular systems. Unlike previous works, our method is applied to redshift bins of width 0.2. Comparison between models applied to a wide redshift range versus bin-specific models reveals significant differences in galaxy morphology beyond $z \geq 1$ and a consistent $\sim 25\%$ disagreement. This suggests that using a single model across wide redshift ranges may introduce biases due to the large time intervals involved compared to galaxy evolution timescales. Using the FERENGI code to assess the impact of cosmological effects, we find that flux dimming and smaller angular scales may lead to the misclassification of up to $18\%$ of disk galaxies as spheroids or irregulars. Contrary to previous studies, we find an almost constant fraction of disks ($\sim 60\%$) and spheroids ($\sim 30\%$) across redshifts. We attribute discrepancies with earlier works, which suggest a decreasing fraction of disks beyond $z \sim 1$, to the biases introduced by visual classification. Our claim is further strengthened by the striking agreement to the results reported by Lee et al. (2024) using an objective, unsupervised method applied to James Webb Space Telescope data. Exploring mass dependence, we observe a $\sim 40\%$ increase in the fraction of massive ($M_{\rm stellar} \geq 10^{10.5}{\rm M}_{\odot}$) spheroids with decreasing redshift, well balanced with a decrease of $\sim 20\%$ in the fraction of $M_{\rm stellar} \geq 10^{10.5}{\rm M}_{\odot}$ disks, suggesting that merging massive disk galaxies may form spheroidal systems.