A Conversational Healthcare Companion in Kannada

Abstract

This study presents an AI-powered bilingual healthcare chatbot to enhance accessibility to primary medical assistance by enabling seamless interactions in both Kannada and English—addressing a critical gap in digital healthcare solutions for multilingual populations. Integrating machine learning–based symptom prediction, voice-enabled communication, secure SQLite-driven appointment scheduling, and Gemini AI for natural conversational responses, the system offers a unified and intelligent healthcare support framework. A multi-class classification model covering 41 disease categories was developed using symptom-level inputs derived from a large-scale clinical dataset comprising approximately 4,900 patient records. To ensure robust and unbiased evaluation, 5-fold stratified cross-validation was employed. Experimental results show that the Random Forest–based model achieved an average classification accuracy of 91%, with consistently balanced precision, recall, and F1-scores across disease classes. Additional noise-injection experiments further confirm the model's robustness under realistic symptom uncertainties. These findings highlight the system's effectiveness as a first-level clinical decision support tool. The key novelty of this work lies in the seamless integration of bilingual conversational AI, predictive analytics, and automated appointment management, offering an end-to-end, accessible, and context-aware healthcare assistance platform. This contribution is particularly significant for resource-constrained and linguistically diverse regions, where timely and reliable medical guidance remains a critical challenge.

Ähnliche Arbeiten