Classification models often exhibit intersectional group unfairness due to the underrepresentation of minority intersectional groups, posing a significant challenge in ensuring fair classification for all groups. Intersectional group fairness considers fairness across multiple dimensions of identity, such as race, gender, and socioeconomic status, recognizing that individuals may experience adverse forms of discrimination or disadvantage based on the intersections of these identities. This paper proposes the Intersectional Fair Transfer Learning (IFTL) method to address this issue by introducing a two-phase strategy utilizing balanced synthetic data and transfer learning. In the first phase, a model is trained on a balanced synthetic dataset w.r.t intersectional groups and label values to establish a fair decision boundary. Subsequently, parameter sharing is employed in the second phase to adapt the pre-trained model to real-world data, aiming to improve classification accuracy while maintaining fairness. Experimental results on two depression datasets demonstrate the efficacy of the proposed IFTL method.