This survey provides a comprehensive overview of the state-of-the-art methods in human activity recognition (HAR), encompassing both classical machine learning techniques and recent advancements. It investigates a wide range of approaches that leverage diverse input modalities, including accelerometer data, video sequences, and audio signals. To navigate the vast and growing HAR literature, the survey introduces a novel methodology using large language models (LLMs) to efficiently filter and pinpoint relevant academic papers, reducing manual effort and ensuring the inclusion of influential works. The survey also provides a taxonomy of the examined literature to enable scholars to access and study HAR approaches rapidly and systematically. The introduction highlights the importance of HAR in various applications, such as health monitoring, smart homes, security surveillance, sports analytics, and human-robot interaction. It outlines the systematic identification and classification of activities based on sensor-derived data, emphasizing the role of wearable devices and the challenges in data preprocessing and feature extraction. The survey then discusses the evolution of HAR, focusing on the integration of deep learning techniques, particularly convolutional neural networks (CNNs) and recurrent neural networks (RNNs), which have enhanced the accuracy and complexity of HAR systems. The methodology section details the data sources, natural language processing techniques, and data compilation process. The authors used IEEE, MDPI, and arXiv repositories to gather data, employing NLP to extract keywords and assess paper quality. The taxonomy method section outlines the metrics used to classify the collected literature, including computational cost, approach, performance, datasets, and supervision. The budget section provides financial and computational costs associated with the research. The devices and processing algorithms section delves into the sensors and algorithms used in HAR. It categorizes sensors into body-worn, object, ambient, and hybrid sensors, and discusses the algorithms, focusing on deep neural networks (DNNs) such as CNNs and RNNs. The section also highlights the advantages and considerations of using these models in HAR. The sensor-based HAR section explores the applications of HAR using accelerometer and IMU signals, WiFi, and radar imaging, as well as methods that combine multiple modalities. It presents several studies that leverage these sensors for activity recognition, demonstrating the transformative impact of sensor fusion on HAR's landscape.This survey provides a comprehensive overview of the state-of-the-art methods in human activity recognition (HAR), encompassing both classical machine learning techniques and recent advancements. It investigates a wide range of approaches that leverage diverse input modalities, including accelerometer data, video sequences, and audio signals. To navigate the vast and growing HAR literature, the survey introduces a novel methodology using large language models (LLMs) to efficiently filter and pinpoint relevant academic papers, reducing manual effort and ensuring the inclusion of influential works. The survey also provides a taxonomy of the examined literature to enable scholars to access and study HAR approaches rapidly and systematically. The introduction highlights the importance of HAR in various applications, such as health monitoring, smart homes, security surveillance, sports analytics, and human-robot interaction. It outlines the systematic identification and classification of activities based on sensor-derived data, emphasizing the role of wearable devices and the challenges in data preprocessing and feature extraction. The survey then discusses the evolution of HAR, focusing on the integration of deep learning techniques, particularly convolutional neural networks (CNNs) and recurrent neural networks (RNNs), which have enhanced the accuracy and complexity of HAR systems. The methodology section details the data sources, natural language processing techniques, and data compilation process. The authors used IEEE, MDPI, and arXiv repositories to gather data, employing NLP to extract keywords and assess paper quality. The taxonomy method section outlines the metrics used to classify the collected literature, including computational cost, approach, performance, datasets, and supervision. The budget section provides financial and computational costs associated with the research. The devices and processing algorithms section delves into the sensors and algorithms used in HAR. It categorizes sensors into body-worn, object, ambient, and hybrid sensors, and discusses the algorithms, focusing on deep neural networks (DNNs) such as CNNs and RNNs. The section also highlights the advantages and considerations of using these models in HAR. The sensor-based HAR section explores the applications of HAR using accelerometer and IMU signals, WiFi, and radar imaging, as well as methods that combine multiple modalities. It presents several studies that leverage these sensors for activity recognition, demonstrating the transformative impact of sensor fusion on HAR's landscape.