Currently, the government is expanding policies and support to promote smart farms. To ensure the reliability of smart farm sensor data, a system capable of promptly detecting abnormal sensor values is essential. Accordingly, this paper proposes a system for detecting anomalies in sensor data and compares the performance of several machine learning-based anomaly detection algorithms to identify the optimal model for reliable data acquisition. The algorithms evaluated include One-Class SVM, Isolation Forest, Local Outlier Factor (LOF), and a Variance-Based Method. The analysis was conducted using temperature and humidity data collected from a tomato smart farm between January 1, 2020, and December 31, 2021. As a result, the LOF model demonstrated the highest overall performance in both quantitative metrics (e.g., F1-score) and visual anomaly distribution consistency. In particular, the LOF model was effective in capturing local density variations and detecting subtle anomalies across continuous time segments. These findings suggest that the LOF algorithm is the most suitable model among those tested for reliable anomaly detection in smart farm sensor environments. The proposed system can contribute to preventing equipment malfunctions and crop damage by enhancing the robustness of environmental monitoring in smart farms.