As is well known, the fiber–matrix interface in composite materials plays a crucial role in determining the overall mechanical performance of carbon fiber–reinforced plastics (CFRP), and numerous methods for evaluating interfacial properties have been proposed and implemented over the years [1]. This study adopts a pinhole pull-out test focusing on a single carbon fiber to evaluate the interface, probably with the highest success rate; however, experimentally derived interfacial strength is often underestimated as an apparent value because it assumes uniform interfacial shear stress along the embedded length, whereas the actual interface experiences nonuniform shear together with tensile stress. Although finite element method (FEM) can estimate the true interfacial strength, repeatedly performing FEM for each change in material parameters is inefficient. To address this issue, we propose a workflow in which FEM simulations covering a wide range of material properties and test conditions are conducted in advance to construct a database, which is then organized by machine learning to enable rapid prediction. Fundamental data were generated using Abaqus 2020 with an axisymmetric pinhole pull-out model, where the resin was modeled as an elastoplastic material and the interface was represented by cohesive elements. Interfacial strength, resin properties, fiber diameter, and embedment length were systematically varied, and design of experiments reduced 3,240 combinations to approximately one-third with balanced coverage. Dataset construction was further streamlined by an automated application that generates inp files, executes Abaqus analyses, and extracts fiber-stress metrics from odb files without GUI operations. A regression model using AutoFeat, Optuna-tuned CatBoost, and ensemble learning was deployed as a Streamlit web application for instant prediction. For 54 training samples, all predictions were within 5% relative error, and for 120 validation samples, 119 cases were within 10%, demonstrating high predictive accuracy across a wide range of conditions.