Fingering decisions play a critical role in classical guitar performance, balancing technical ease and musical expression. While existing computational models focus on other guitar types, they often fail to meet the specific needs of classical guitar. Therefore, we addressed this gap by introducing a comprehensive fingering dataset and developing a prediction model tailored to classical guitar. A dataset of 40 annotated etudes was created, covering a wide range of technical and stylistic challenges. The fingering prediction model was constructed using an ensemble approach, first predicting the string and then the specific fingering, mimicking the decision-making process of classical guitarists. The model achieved high accuracy, with 0.944 for string prediction and 0.903 for fingering prediction. This research contributes a valuable tool for both pedagogical and performance purposes, improving fingering decisions by combining technical optimization with musical interpretation, and offering a robust foundation for future studies in classical guitar fingering.