We study the phenomenological signatures associated with a light fermiophobic Higgs boson within the type-I two-Higgs-doublet model at the HL-LHC. Our exhaustive parameter scan revealed a captivating mass range between 1 GeV and 10 GeV. This range retains a substantial number of viable parameter points, primarily due to the current experimental difficulties in probing soft decay products of the light fermiophoibic Higgs, two photons. A major obstacle arises as two photons from the $\hf$ decay tend to merge into one jet because of their proximity. This leads to dominating QCD backgrounds. To address this, we utilize EFlow objects within the Delphes framework, identifying a jet containing two photons, termed a diphoton jet. Through our full detector-level simulations across 18 benchmark points, the majority presented signal significances beyond 5 at an integrated luminosity of 3/ab. In challenging scenarios with a heavier charged Higgs boson, our incorporation of machine learning techniques demonstrated a significant enhancement.