This study explores a new approach to guiding art education. The evaluation of drawing pictures mainly uses subjective evaluation, which has the problem that the results vary from rater to rater. We propose a linear regression evaluation model that predicts subjective evaluation scores for pencil still-life drawings based on extracted image features. The model estimates subjective evaluations of composition, shape, lightness/darkness, sense of space, and overall strength. Extending our previous study, we enhanced the model to generate personalized feedback for students by analyzing the feature values that contribute to the predicted evaluation scores. The feedback offered suggestions for areas of improvement, such as adjusting the composition, refining shapes, and enhancing contrast. Experiments conducted on a dataset of 176 still-life drawings yielded a multiple correlation coefficient of 0.86 between the predicted and actual subjective evaluation scores, indicating the potential effectiveness of our approach. Additionally, a user study with design teachers suggested that the generated feedback was relevant and could help improve their drawing skills. By combining the prediction of subjective evaluations with targeted feedback generation, our system presents a promising tool for supporting art education and potentially enhancing students’ learning experience.