OLIY TA’LIM MUASSASALARIDA AKADEMIK KADRLAR QO‘NIMSIZLIGINI ERTA OGOHLANTIRISH TIZIMINING TASHKILIY-USLUBIY MODELI

The article develops an organizational and methodological model of an early warning system that anticipates high turnover intention among academic staff of higher education institutions. Using data from an anonymous survey of 1,005 faculty members, attributes recorded in a staff register and indicators from a periodic survey were compared with simple and penalized logistic regression, random forest and gradient boosting. Models based on register attributes alone showed low predictive quality (AUC from 0.628 to 0.684), whereas adding survey indicators raised it to 0.896 and increased the share of at-risk employees identified from 44.8 to 80.4 percent. Predictive quality is determined by the composition of the data rather than by the complexity of the algorithm. On this basis, a six-stage operating cycle is proposed that combines the staff register, an anonymous survey twice a year, aggregation of results by organizational unit and a final decision on any employee taken by a human