The goal of antiviral treatment for chronic hepatitis B is to prevent disease progression by suppressing hepatitis activity; 48-week treatment with peg-IFN is used as first-line antiviral treatment for persistently infected individuals. However, because treatment efficacy is heterogeneous among individuals and treatment has side effects, it is desirable to be able to identify early which individuals would benefit from treatment. In this study, time series data on three serum biomarkers, HBV DNA, HBsAg and HBcrAg, which determine treatment response, were used to stratify the treatment efficacy of patients. First, a mathematical model describing the time-series dynamics of these biomarkers was constructed and parameters were estimated for each individual using non-linear mixed effects modeling. Clustering of patients based on these parameters showed that patients with high treatment efficacy were concentrated in a specific cluster. This cluster was characterized by lower biomarker levels at baseline compared to other clusters, with HBsAg and HBV DNA declining by more than 1 log10 during the first several weeks of treatment. The degree of decline in the amount of cccDNA remaining in hepatocytes was also greater. Therefore, a machine learning model was created to predict this cluster using random forest. The results showed that using both the initial biomarker levels at the start of treatment and the cumulative levels up to several weeks after treatment, it was possible to identify a group of patients with high treatment efficacy with sufficient accuracy. Other blood markers and patient background factors were also found to be associated with treatment response. Thus, dynamical systems phenotyping based on multivariable time-series biomarkers allows patient stratification and prediction of treatment efficacy. We expect that such a method could also be used to stratify patients with other diseases.