نوع مقاله : مقاله پژوهشی
نویسندگان
1 گروه حسابداری، واحد ارومیه، دانشگاه آزاد اسلامی، ارومیه، ایران
2 گروه حسابداری، واحد ارومیه، دانشگاه آزاد اسلامی ارومیه، ایران
3 دانشیار، گروه حسابداری، واحد ارومیه، دانشگاه آزاد اسلامی ارومیه، ایران
4 دانشیار گروه حسابداری، دانشکده اقتصاد و مدیریت، دانشگاه ارومیه، ارومیه، ایران
چکیده
کلیدواژهها
موضوعات
عنوان مقاله [English]
نویسندگان [English]
Topic and Objective: This research aimed to design and validate a comprehensive machine learning-based model for combating financial corruption in Iran's tax system. Tax corruption poses a significant challenge to the Iranian economy by exacerbating economic inequality, reducing public government revenues, and obstructing sustainable development.
Methodology: The study is applied in purpose and descriptive-analytical in nature. The statistical population consisted of employees and experts from the Iranian National Tax Administration, with 140 participants selected through stratified random sampling. Data were collected using a researcher-developed questionnaire comprising 71 items across nine dimensions. Analysis employed advanced machine learning algorithms, including Random Forest, Gradient Boosting, and Deep Neural Networks. To enhance interpretability, SHAP values, Partial Dependence Plots (PDP), K-means clustering, and Fuzzy Analytic Hierarchy Process (AHP) were applied.
Findings: Results revealed that complexity of tax laws, lack of process transparency, and weak oversight were the most influential factors in tax corruption. Cluster analysis identified three tax system profiles: traditional, transitional, and digital. Prioritization of solutions highlighted law clarification and simplification, education and cultural promotion, and accountability enhancement as the most effective measures.
Conclusion, Originality, and Contribution to Knowledge: Scenario simulations indicate that full implementation of the proposed strategies could reduce tax corruption by up to 47.8% over five years. The study's originality stems from its advanced integration of machine learning with interpretable methods and multi-criteria decision-making, delivering an indigenous, comprehensive, and prioritized model. This framework offers a solid scientific basis for anti-corruption policymaking and future research.
کلیدواژهها [English]
)