<p>Recognizing Human lower limb activity is crucial for many applications, including robotics, rehabilitation, and prosthetics. This study presents a paradigm for predicting real-time lower limb activity using Various Deep Learning (DL) models which include Convolutional neural networks (CNN), Recurrent Neural Networks (RNN) and Long Short-Term Memory (LSTM) networks. Existing data acquisition systems are complex and highly expensive, this work aims to provide a simplified, small, and cost-effective system for data collection. The designed system consists of MyoWare sensor with Arduino uno for data collection with sEMG electrodes at three muscle locations- Femoris (RF), Biceps Femoris (BF), and Vastus Medialis (VM). Data was collected form 5 subjects performing 3 different activities such as sitting, standing and walking. The data collected was preprocessed using band pass and notch filter and it was given to different DL models for predicting activities. The DL models using CNN, RNN and LSTM needed to provide precise and timely predictions of lower limb movements based on real-time data collected by wearable sensors. Precision, recall, F1-score, and accuracy measures were then used to evaluate the framework’s performance to show that it is highly effective at forecasting lower limb activities. The LSTM model delivered excellent results, with an average accuracy of 98.26%, precision of 98.24%, and recall of 98.77%. Additionally, class-wise accuracy for individual subjects demonstrated great performance, with mean accuracies of 97.09% for Sitting, 98.44% for Standing, and 98.6% for Walking. These findings demonstrate the model’s robustness and dependability in appropriately identifying various physical activities.</p>