Comparing Different Machine Learning Architectures for Classifying Medical Terms in Colombian Sign Language

Sign language recognition has been widely studied, demonstrating the feasibility of real-time, high-accuracy inter-pretation of sign-based communication. However, most existing approaches rely on specialized hardware such as Leap Motion Controllers or sensor-equipped gloves, which, despite their effectiveness, limit accessibility. This study aims to contribute to the development of accessible and efficient sign language interpretation systems, particularly to assist medical personnel in understanding their patients, by employing MediaPipe to capture hand motion data from videos of fluent Colombian Sign Language (LSC) users. These features are then structured and used to train and compare various machine learning models, including a Multi-Layer Perceptron (MLP), a Long Short-Term Memory (LSTM) network, and an Enhanced LSTM with attention (ELSTM), with the goal of accurately classifying medical signs and paving the way for future advancements in sign language-based communication tools in Colombia.