Human Activity Recognition (HAR) has gained increasing attention in recent years due to its wide range of applications in health, sports, and surveillance. The ubiquity and unobtrusive nature of inertial measurement unit (IMU) sensors in consumer devices have contributed to the growing popularity of IMU-based HAR systems. However, many existing models lack transparency in their decision-making processes—an essential aspect for both users and health professionals. In this work, we propose a human activity classifier based on the Kalman filter, designed to be interpretable and efficient using IMU data. We evaluate its performance on standard HAR datasets and compare it against classical benchmark classifiers such as logistic regression and support vector machines. Experimental results demonstrate that the proposed model offers a viable and competitive alternative for classifying human activities.