In modern interstitial-free (IF) steel continuous casting processes, the stability of the mold level plays a critical role in determining slab quality. Excessive fluctuations may cause slag entrapment into molten steel, resulting in inclusion defects that degrade the purity and final mechanical properties of the steel. While traditional process monitoring primarily depends on physical experiments and finite element simulations, the advancement of industrial data acquisition technologies has enabled data-driven approaches to demonstrate significant potential for process optimization. Industrial IF steel casting operations involving the collection of mold level fluctuation data, slag entrapment records, and associated process parameters were investigated. Through integrated statistical analysis, time series characterization and machine learning techniques, the intrinsic relationship between mold-level fluctuation patterns and slag entrapment mechanisms was systematically explored, and corresponding control strategies were proposed. The results demonstrate significant correlations between the characteristic parameters of level fluctuations and slag entrapment defects. The implementation of optimized control strategies effectively reduces the probability of inclusion formation.