Open Access

Zaid Saleem Syed

• Anomaly detection is an important aspect of beehive monitoring, enabling beekeepers to take timely action and preventing economic losses caused by the decline of the bee population. Most existing beehive monitoring systems lack machine learning (ML) capabilities for anomaly detection and rely on threshold-based methods or expensive, complex designs. This thesis utilized the DigiBee prototype, a cost- effective monitoring system, to develop an ML-based anomaly detection model using Isolation Forest. Data on temperature, humidity, sound, vibration, and weight from four active beehives was collected between May 1, 2024, and July 25, 2024, alongside environmental data—temperature, relative humidity, and precipitation—from nearby meteorological stations. Exploratory data analysis revealed several limitations: uniform outputs from vibration sensors, calibration errors in weight sensors, and multiple gaps during the data collection period. Correlation analysis indicated a limited influence of external weather variables on internal beehive conditions, while internal sensor parameters displayed stronger correlations with each other. Five models were developed, using either beehive data alone or a combination of beehive and weather data. Results showed that beehive-specific models achieved higher accuracy in detecting localized anomalies, whereas models combining data from multiple beehives generalized better but underrepresented hive-specific issues. Models incorporating weather parameters, especially precipitation, introduced noise and unnecessary dimensions, which, when excluded, improved the model’s prediction by focusing on hive-specific patterns and ensuring anomaly detection remained contextually relevant.