The recent years witnessed a surge in the intellectual interest of statistical methods for personalized dosing. It is important to investigate the Individual Dose Interval (IDI) both for medical applications and statistical researches. In this dissertation, a new method based on Outcome-Weighted-Learning (OWL) is proposed to determine two types of IDIs with different properties: Probability Dose Interval (PDI) and Expectation Dose Interval (EDI). The former identifies a range of doses where the potential outcome is greater than a threshold with a certain probability and the latter identifies a range of doses where the expectation of the potential outcome is greater than a threshold. The method adopts Reproducing Kernel Hilbert Space (RKHS) estimators from non-convex loss functions, which are solved by the Difference-of-Convex (DC) algorithm. The Fish consistency and the convergence rates of the estimators are discussed. Numerical simulations show an advantage of the proposed method over methods based on traditional outcome modeling. In addition, this new method is applied to determine personalized Hemoglobin A1c (HbA1c) control intervals for diabetic patients.