Thesis:

Investigating people's behavior in a smart home using sensors and algorithms XGBoost٫ Random Forest٫ Nearest Neighbor٫ and Hybrid Algorithms

With the rapid expansion of emerging technologies and the increasing deployment of smart homes, recognizing and analyzing human behavior in such environments has become a central research focus. Sensor data collected in smart homes contain large volumes of information about residents’ daily activities but are often accompanied by challenges such as noise, missing values, and class imbalance. The main objective of this study is to propose an effective framework for accurately recognizing human activities and identifying rare and potentially critical behaviors. The proposed framework begins with extracting and structuring raw sensor events, followed by preprocessing and feature engineering to prepare the data for analysis. Subsequently, machine learning models are trained and evaluated using standard performance metrics including Accuracy, Precision, Recall, F1-score, and AUC. The architecture integrates the XGBoost algorithm with two ensemble strategies, Soft Voting and Stacking, aiming to increase classification accuracy and enhance the ability to detect rare behaviors. The results demonstrate that the proposed framework is capable of analyzing real-world data and providing practical solutions for improving residents’ quality of life.

Keywords: Smart Home, Human Behavior, Machine Learning, Sensors, Data Analysis, Hybrid Algorithm