Chronic　mental　pressure　affects　human　health　directly　by　causing　a　series　of　pathological　and　physiological　risks.　Effective　methods　of　evaluating　psychological　pressure　can　detect　and　assess　real-time　stress　states,　warning　people　to　pay　close　attention　to　their　health.　Focusing　on　stress　assessment,　this　study　improved　the　support　vector　machine　(SVM)　algorithm　to　assess　the　stress　state　via　surface　electromyographic　signals.　After　the　samples　were　clustered,　the　cluster　results　were　given　to　the　loss　function　of　SVM　to　screen　training　samples.　With　the　imbalance　problem　after　screening,　a　weight　was　given　to　the　loss　function　to　reduce　the　prediction　tendentiousness　of　the　classifier,　thereby　decreasing　the　error　of　the　training　sample　and　compensating　for　the　influence　of　the　unbalanced　samples.　This　improved　algorithm　increased　the　classification　accuracy　from　68%　to　79%　and　reduced　the　running　time　from　2026.5　s　to　541.3　s.　Experimental　results　show　that　this　algorithm　can　effectively　avoid　the　influence　of　individual　differences　on　stress　appraisal　effect　and　reduce　the　computational　complexity　during　the　training　phase　of　the　classifier.