In today’s automated systems, large amounts of log data are generated every second. These logs contain important information about system behavior, commands executed, and operational parameters. While logs help in monitoring performance, they can also become a target for unauthorized modifications or malicious tampering. Detecting such abnormal behavior manually is extremely difficult, especially when the dataset is large. This project presents a Log-Based Anomaly Detection System that uses machine learning techniques to automatically identify abnormal patterns in structured log data. The system analyzes command-related features, coordinate values, and execution parameters to determine whether a log entry is normal or anomalous. To achieve this, models such as XGBoost and a Feedforward Deep Neural Network were implemented and compared. The dataset contains hundreds of thousands of log records, making it suitable for building a robust classification system. The trained models achieved very high accuracy and strong evaluation metrics, showing that machine learning can effectively detect subtle anomalies in log data. The project also includes visualization of results and a risk scoring mechanism to better interpret the severity of detected anomalies. Overall, this system demonstrates how data-driven approaches can improve security and reliability in automated environments.