Abstract
Human Activity Recognition (HAR) technology has received significant attention in recent years for its potential to improve applications in various fields, including human-computer interaction, autonomous driving, disease diagnosis, healthcare, and sports. This technology focuses on collecting and analyzing data from sensors embedded in smartphones and wearable devices, which provide real-time insights into different individual behaviors. With advances in machine learning techniques, it is now possible to analyze this complex data with high efficiency and ease, enabling the development of intelligent systems capable of automatically recognizing human activities. This study investigates the performance of several classic machine learning models, including Logistic Regression (LR), Support Vector Machines (SVM), and Decision Trees (DT), to classify diverse human activities. The hyperparameters of each model were optimized using Cross Validation (GridSearchCV) to achieve the best system performance. The benchmark used to train and test the models was the HAR dataset from Kaggle, which includes labeled data for different human activities such as walking, sitting, standing, and climbing and descending stairs. The results showed that the Support Vector Machines model outperformed other algorithms, achieving an accuracy of 96.67%. Furthermore, the use of GridSearchCV significantly enhanced.