Development and validation of a deep learning–based assessment tool for teacher leadership: A case study from Xinjiang, China

Abstract

Teacher leadership is widely regarded as a critical driver of school reform and educational quality improvement. Although the field has been extensively studied, empirical research remains limited in Xinjiang, China—a region characterized by its multiethnic and multilingual context. To address this gap, the present study developed and validated a culturally sensitive assessment tool based on a sample of 371 primary and secondary school teachers from Xinjiang. A structured questionnaire was designed encompassing four dimensions: professional guidance, educational collaboration, cross-cultural ICT-based teaching competence, and leadership cognition. In addition, we introduced an interpretable deep learning model—ITL-LSTM (Interpretable Teacher Leadership LSTM)—which employs a Diagonal BiLSTM structure for dynamic classification of teacher leadership profiles, achieving a prediction accuracy of 90.10%. The findings indicate that the proposed tool demonstrates strong applicability and scalability within the Xinjiang context, providing effective support for dynamic evaluation, personalized development, and evidence-based decision-making in multicultural educational settings.