Towards Enhanced Gym Safety and Efficiency: CNN-LSTM Models for Identifying Leg Builder Workouts

The algorithm formulates a hybrid deep learning structure that utilizes the Convolutional layer with the LSTM network. The model underwent training and validation using a dataset consisting of 2040 photos that depict five specific exercise actions: body-weight squats, lunges, lying leg presses or hamstring curls, and calf lifts with a weight plate. In our study, we’ve found that the participants have excellent classification skills with precision scores in the range from to $97.21 \%$; and the F1-score varied from $91.53 \%$ to $\mathbf{9 4 . 9 2 \%}$. The accuracy of each class was from $92 \%$ to $97 \%$, with high accuracy noted. The confusion matrix offers a comprehensive view for the review of rightly confirmed positives with the highest level of accuracy found in Classes 4 and 5. However, difficulties in differentiation with Classes 3 and 4 surfaces, thus, become a convergent feature representation that needs to be a subject of future study. The convolutional net, coupled with extended short-term networks, showed the capability to reflect fitness modes in both spatial and time-related movement, resulting in superior precision compared to other sports and rehab application methods. The management capability of such an approach reveals that utilizing powerful machine learning tools is an efficient means of promoting an increase in the number of hours available for training and preventing fatal accidents through immaculate compliance to workouts.