The paper presents a novel facial expression recognition (FER) system that combines the use of k-nearest neighbors (KNN) and long short-term memory (LSTM) algorithms to enhance efficiency and accuracy. The system consists of two primary steps: feature extraction and classification. For feature extraction, the Local Binary Patterns (LBP) algorithm is used to capture texture information from facial images. In the classification stage, KNN and LSTM are employed to recognize facial expressions. KNN is effective for finding the k closest neighbors to a test sample, while LSTM captures temporal relationships in time series data, addressing KNN's limitations in handling temporal information. The system is evaluated on the CK+ and Oulu-CASIA datasets, achieving state-of-the-art performance in terms of F1-score and precision. The proposed system outperforms other state-of-the-art methods, including deep learning systems, and has potential applications in human-computer interaction, emotion detection, and behavior analysis. The methodology involves data collection, pre-processing, feature extraction, model selection, training, evaluation, and deployment. The results demonstrate high accuracy and robustness, making the system a promising tool for various real-world applications.The paper presents a novel facial expression recognition (FER) system that combines the use of k-nearest neighbors (KNN) and long short-term memory (LSTM) algorithms to enhance efficiency and accuracy. The system consists of two primary steps: feature extraction and classification. For feature extraction, the Local Binary Patterns (LBP) algorithm is used to capture texture information from facial images. In the classification stage, KNN and LSTM are employed to recognize facial expressions. KNN is effective for finding the k closest neighbors to a test sample, while LSTM captures temporal relationships in time series data, addressing KNN's limitations in handling temporal information. The system is evaluated on the CK+ and Oulu-CASIA datasets, achieving state-of-the-art performance in terms of F1-score and precision. The proposed system outperforms other state-of-the-art methods, including deep learning systems, and has potential applications in human-computer interaction, emotion detection, and behavior analysis. The methodology involves data collection, pre-processing, feature extraction, model selection, training, evaluation, and deployment. The results demonstrate high accuracy and robustness, making the system a promising tool for various real-world applications.