Data　balance　has　a　great　impact　on　the　performance　of　neural　network　training　model,　and　the　model　trained　with　high-quality　data　is　often　more　robust.　In　the　task　of　malicious　code　detection　and　classification,　the　number　of　samples　is　often　very　different　among　different　malicious　families.　If　the　raw　data　is　directly　used　for　model　training,　it　often　brings　over　fitting　problems.　Aiming　at　this,　this　paper　first　analyzes　the　common　data　equalization　methods　in　the　classical　machine　learning　model　and　neural　network　model,　and　then　designs　a　hybrid　dynamic　sampling　strategy　based　on　swarm　intelligence　optimization　algorithm　to　improve　the　performance　of　the　model.　Experimental　results　show　that　our　malicious　code　data　balancing　strategy　is　effective.　©　2022,　Springer　Nature　Singapore　Pte　Ltd.