To　investigate　the　axial　compressive　properties　and　failure　modes　of　18　locally　corroded　square　steel　tube　short　columns,　salt　spray　corrosion　tests　were　conducted.　The　findings　indicate　that　local　corrosion　in　the　horizontal　direction　significantly　impacts　the　ultimate　bearing　capacity　of　the　specimens,　with　all　failures　occurring　in　the　corroded　areas.　However,　the　influence　of　the　corrosion　area　on　the　ultimate　bearing　capacity　and　ductility　of　the　specimens　is　not　evident.　A　local　corrosion　model　for　square　steel　tube　short　columns　was　developed　using　the　finite　element　method.　The　ultimate　bearing　capacity　determined　through　finite　element　simulation　closely　aligns　with　experimental　results,　validating　the　reliability　of　the　simulation　approach.　Moreover,　a　comparison　is　made　between　the　prediction　models　of　the　BP　neural　network　and　the　GA-BP　neural　network,　with　the　prediction　effectiveness　of　each　model　evaluated　using　the　leave-one-out　method.　The　verification　results　indicate　that　the　GA-BP　neural　network　model　demonstrates　superior　predictive　performance,　with　an　average　error　of　3.14%.