Named Entity Recognition (NER) is a critical task in Natural Language Processing (NLP) with uses in information extraction, machine translation, search engine, document summarization, sentiment analysis, language comprehension, and question-answering. Bodo is a low-resource language that suffers from the lack of annotated corpora and linguistic resources. This paper suggests a deep learning-based method to NER for Bodo using Long Short-Term Memory (LSTM), Gated Recurrent Units (GRUs), and Convolutional Neural Networks (CNNs). Data augmentation and transliteration methods are utilized to overcome data paucity. The results of experiments indicate that CNN performs best compared to other structures with an accuracy of 99.91%, followed by GRU at 99.36% and LSTM at 96.5%. SHAP analysis is also used for feature importance in order to extend model interpretability. This work supports the improvement of NER research for low-resource languages and demonstrates the efficiency of deep learning in processing low-resource linguistic issues.