LSTM Based Next Word Prediction for Workout Descriptions in an Android Mobile Application
Joshua Felix, T. Mathu, M. Roshni Thanka · 2024
Keeping up with workouts at the gym can be hectic, especially when doing a sequence of continuous workout combinations. In order to improve the gym workout experience, this study presents a combination of an LSTM next word prediction model in an android application. The Optimization algorithm used is Adam’s optimizer which is used for training the neural network. Softmax activation is the activation function utilized in the neural network’s output layer. This prediction algorithm is effortlessly integrated into an Android application to provide users with activity descriptions that are updated in real-time and correspond closely with their predetermined fitness goals. This research makes a significant contribution to the rapidly developing field of fitness technology by influencing the convergence of natural language processing approaches with mobile app development. For the dataset, we have used MegaGym dataset from kaggle. The dataset contains 2909 different exercises. We have obtained accuracy as 95% in the final epoch and F1-Score, precision, recall values close to 96%.