Surface Electromyography (sEMG) signals provide valuable insights into muscle activity during physical movements, making them essential for real-time rehabilitation monitoring. This study proposes a computationally efficient Multi-Layer Perceptron (MLP) classifier for classifying elbow joint movements (flexion, extension, and rest) using only two sEMG sensors. Noise removal and feature extraction techniques were applied to pre-process the EMG signals, followed by feature selection using permutation importance with most significant features. A Multi-Layer Perceptron (MLP) classifier trained on a dataset split into 70% training and 30% testing achieved 90% accuracy. The model’s robustness, demonstrated through high precision and recall, ensures reliable movement detection. With a 200ms window and 170ms overlap, the system maintains a 30ms update speed. The model was evaluated using accuracy, precision, recall, and F1-score, demonstrating high reliability, particularly for rehabilitation scenarios. The findings indicate that reducing sensor input while optimizing feature selection can maintain high classification performance.