Amharic Text Based Chatbot for Driving Trainee Using Machine Learning Approach

Abstract

The increasing demand for effective and user-friendly Amharic chatbots, particularly in the context of assisting driver's license trainees, highlights the need for advanced natural language processing (NLP) systems tailored to the Amharic language. This study addresses the challenge of developing a chatbot capable of understanding and responding accurately to user queries in Amharic, a language with a complex writing system and limited resources for NLP research. The research aims to design and implement an Amharic chatbot that can provide accurate responses to driver's license-related queries, thereby improving the learning experience for trainees. The study employs a comprehensive methodology, beginning with the collection and preprocessing of Amharic text data, followed by the application of various text normalization, cleaning, tokenization, and stop-word removal techniques. The vectorized textual data is then used to train a machine learning model using the Keras Sequential API, with an 80/20 split between training and validation datasets. The model architecture includes a deep neural network with three layers, optimized through a series of heuristic adjustments. The study also explores different text representation methods, with a focus on the bag-of-words (BoW) approach, to find the most suitable technique for the limited Amharic text corpus. Experimental scenarios were conducted to evaluate the performance of the model, using accuracy and loss metrics. The chatbot achieved an accuracy of 87.9% on the test dataset, demonstrating its potential effectiveness in real-world applications. TensorBoard was employed to visualize the model's performance over 150 training epochs, revealing the model's strong generalization capabilities. The key findings indicate that the proposed chatbot model is capable of providing accurate and contextually appropriate responses to user queries, making it a valuable tool for license trainees. The study's implications suggest that future research should focus on expanding the Amharic text corpus, exploring more advanced NLP techniques, and refining the model to enhance its conversational flexibility and accuracy. This research contributes to the growing field of Amharic NLP and sets the stage for further advancements in chatbot development for underresourced languages.