This chapter discusses advancements in HAR systems using wearable sensors, covering machine learning, ensemble learning, and deep learning approaches. It highlights the effectiveness of shallow learning models in online recognition, energy-efficient ensemble learning variants, and ensemble methods like random forest for improved accuracy and efficiency. The chapter also explores deep learning techniques, including federated learning, graph neural networks, and domain adaptation, enhancing model generalization. Additionally, BiLSTM-based frameworks and transfer learning strategies optimize real-time predictions and adaptation to new users. These advancements contribute to the development of robust and efficient HAR systems for real-world applications. Researchers around the globe practice HAR system development using various tools and techniques incorporating shallow, ensemble, and deep learning algorithms. Numerous feature engineering and sensor modalities are used for statistical feature generation and raw sensor data collection, respectively. This chapter comprehensively outlines the benchmark state-of-the-art wearable sensor-based HAR systems in multiple sub-sections for profound insights as follows.