Lower Limb Prosthetics Activity Recognition entails identifying and categorizing human movements related to activities using the lower limbs. Still, the detection of lower limb prosthesis activity is hindered by intricate movements, inconsistent user patterns, and challenges in real-time processing. To overcome hurdles, one can use advanced sensors, personalized training, and ensemble learning approaches to improve accuracy. The present study addresses the issue by employing the Optimized CNN-LSTM model, which enhances the accuracy of activity identification in lower limb prostheses. The suggested approach comprises four main steps: pre-processing, feature extraction, feature selection, and classification. The gathered pictures undergo pre-processing using an enhanced Gaussian Bilateral Filter and Complete Ensemble Empirical Mode Decomposition (CEEMD). Features are derived from pre-processed data using a Dual-tree Complex Wavelet Transform (DTCWT). The extracted features are then transferred to the Feature categorization procedure. The classification step is performed using the Optimized CNN-LSTM technique. This model has been improved using the Adapted Bat Algorithm to enhance its performance. After the classification, the feedback level is included in the proposed approach. The suggested model is implemented using the PYTHON programming language, and its performance is assessed using metrics such as accuracy, precision, recall, F-score, specificity, sensitivity, MCC, NPV, FPR, and FNR. Suggested optimized CNN-LSTM achieves higher accuracy (99.23%) and precision (98.83%) compared to current techniques.