This paper presents a systematic mapping study on the application of AI techniques in requirements classification. The study analyzed 61 studies published between 2012 and 2022, focusing on the use of AI in identifying and classifying software requirements. The findings indicate that transfer learning-based approaches are extensively used and yield the most accurate results compared to traditional machine learning (ML) and deep learning (DL) techniques. Support Vector Machine (SVM) and Convolutional Neural Network (CNN) are widely used in the selected studies. Precision and recall are the most commonly used metrics for evaluating the performance of automated techniques. While these AI approaches show promising results in classification, their applicability in complex and real-world settings remains underexplored. The study highlights the need for closer collaboration between requirements engineering (RE) and AI techniques to address open issues in the development of real-world automated systems. The study also identifies key findings, limitations, and open challenges in the field, including the need for new benchmark datasets, computational cost issues, and the scarcity of evaluation and reported results. The results show that pre-trained language models like BERT, RoBERT, and DistilBERT perform well in requirements classification, achieving accuracy above 80%. The study emphasizes the importance of using appropriate evaluation metrics and the need for further research to improve the generalization of AI techniques in real-world scenarios.This paper presents a systematic mapping study on the application of AI techniques in requirements classification. The study analyzed 61 studies published between 2012 and 2022, focusing on the use of AI in identifying and classifying software requirements. The findings indicate that transfer learning-based approaches are extensively used and yield the most accurate results compared to traditional machine learning (ML) and deep learning (DL) techniques. Support Vector Machine (SVM) and Convolutional Neural Network (CNN) are widely used in the selected studies. Precision and recall are the most commonly used metrics for evaluating the performance of automated techniques. While these AI approaches show promising results in classification, their applicability in complex and real-world settings remains underexplored. The study highlights the need for closer collaboration between requirements engineering (RE) and AI techniques to address open issues in the development of real-world automated systems. The study also identifies key findings, limitations, and open challenges in the field, including the need for new benchmark datasets, computational cost issues, and the scarcity of evaluation and reported results. The results show that pre-trained language models like BERT, RoBERT, and DistilBERT perform well in requirements classification, achieving accuracy above 80%. The study emphasizes the importance of using appropriate evaluation metrics and the need for further research to improve the generalization of AI techniques in real-world scenarios.