Feature　selection　is　a　method　used　to　decrease　the　number　of　features　by　removing　unwanted,　noisy　and　inconsistent　data　while　maintaining　classification　accuracy.　Most　researchers　have　focused　on　using　metaheuristic　algorithms　to　select　the　best　possible　features　to　improve　and　simplify　the　dataset　quality.　However,　the　traditional　optimization　method　tends　to　suffer　from　local　optimality　problems　as　the　increasing　of　features　in　datasets.　In　this　paper,　an　investigation　is　conducted　to　assess　the　performance　of　flow　direction　algorithm　(FDA)　in　enhancing　the　classification　accuracy　by　performing　feature　selection.　Eight　datasets　obtained　from　UCI　machine　learning　repository　are　used　to　perform　comparative　studies　with　existing　algorithms　known　as　differential　evolution　(DE),　biogeography-based　learning　particle　swarm　optimization　(BLPSO),　henry　gas　solubility　optimization　(HGSO)　and　African　vulture　optimization　algorithm　(AVOA).　The　results　reported　that　FDA　obtains　best　mean　accuracy　in　the　comparative　studies　with　the　selected　algorithms　and　the　number　of　features　selected.　©　2023,　The　Author(s),　under　exclusive　license　to　Springer　Nature　Singapore　Pte　Ltd.