This paper presents a biosensor-driven IoT wearable system designed to accurately track and localize human movements. The system leverages smartphone sensors, including accelerometers, gyroscopes, and GPS, to recognize both physical activities and location-based patterns. Key contributions include: 1. **Denoising and Preprocessing**: Raw sensor data is preprocessed using Butterworth and median filters to reduce noise and enhance signal clarity. 2. **Feature Extraction and Selection**: Features are extracted from inertial and GPS data using methods like Shannon entropy, linear prediction cepstral coefficients (LPCCs), skewness, kurtosis, and Mel-Frequency Cepstral Coefficients (MFCCs). Variance thresholding is used to select significant features. 3. **Data Augmentation**: Permutation-based data augmentation is applied to address class imbalance issues. 4. **Feature Optimization**: The Yeo–Johnson power transformation is used to optimize feature distributions, improving model performance. 5. **Multi-Layer Perceptron (MLP) Classification**: The classified features are fed into an MLP for activity recognition, achieving high accuracy rates. The system was evaluated using the Extrasensory and Sussex Huawei Locomotion (SHL) datasets, achieving 96% and 94% accuracy for physical activities, and 94% and 91% for location-based activities, outperforming existing methods. The study highlights the system's potential in health monitoring, urban navigation, and other IoT applications, while also discussing limitations such as GPS inaccuracies in certain environments.This paper presents a biosensor-driven IoT wearable system designed to accurately track and localize human movements. The system leverages smartphone sensors, including accelerometers, gyroscopes, and GPS, to recognize both physical activities and location-based patterns. Key contributions include: 1. **Denoising and Preprocessing**: Raw sensor data is preprocessed using Butterworth and median filters to reduce noise and enhance signal clarity. 2. **Feature Extraction and Selection**: Features are extracted from inertial and GPS data using methods like Shannon entropy, linear prediction cepstral coefficients (LPCCs), skewness, kurtosis, and Mel-Frequency Cepstral Coefficients (MFCCs). Variance thresholding is used to select significant features. 3. **Data Augmentation**: Permutation-based data augmentation is applied to address class imbalance issues. 4. **Feature Optimization**: The Yeo–Johnson power transformation is used to optimize feature distributions, improving model performance. 5. **Multi-Layer Perceptron (MLP) Classification**: The classified features are fed into an MLP for activity recognition, achieving high accuracy rates. The system was evaluated using the Extrasensory and Sussex Huawei Locomotion (SHL) datasets, achieving 96% and 94% accuracy for physical activities, and 94% and 91% for location-based activities, outperforming existing methods. The study highlights the system's potential in health monitoring, urban navigation, and other IoT applications, while also discussing limitations such as GPS inaccuracies in certain environments.