hassen8/Language_Detection

99%-accurate text classifier for Oromo, Somali, and Swahili — low-resource , African NLP # Language Detection — Oromo, Somali & Swahili > A text classifier for three East African languages achieving **99% accuracy**, built to address the scarcity of NLP tooling for low-resource African languages. --- ## Overview Most language detection tools perform well on European languages but fail on low-resource African languages due to a lack of training data and linguistic research. This project targets three East African languages — **Oromo**, **Somali**, and **Swahili** — building a classifier that can be used as a pre-processing step in downstream NLP pipelines. --- ## Results | Metric | Score | |--------|-------| | Accuracy | **99%** | | Languages | Oromo, Somali, Swahili | --- ## Stack --- ## Usage ### Run the notebook ```bash git clone github.com cd Language_Detection pip install scikit-learn jupyter jupyter notebook Language_detection.ipynb ``` ### Use the saved model ```python import pickle with open("detection_model.pkl", "rb") as f: model = pickle.load(f) texts = ["Akkam jirta?", "Sidee tahay?", "Habari yako?"] predictions = model.predict(texts) print(predictions) # ['oromo', 'somali', 'swahili'] ``` --- ## Files | File | Description | |------|-------------| | `Language_detection.ipynb` | Full training pipeline and evaluation | | `detection_model.pkl` | Serialized trained model | --- ## Context Oromo, Somali, and Swahili collectively have hundreds of millions of speakers yet remain significantly under-represented in mainstream NLP research. This project is part of a broader interest in building language tools for under-resourced communities. --- ## Related Work - Amharic ATS — IEEE 2023 - Ho Language ATS — IEEE 2023