Investigation of machine learning methods to detect the absence of queen bee in the hive

The queen bee is vital for maintaining order and honey production in a honey bee colony. The hive's organization and honey production suffer if the queen bee is absent or becomes unable to perform her duties. Therefore, it is important to monitor the queen bee's presence in the hive. Recent studies have shown promising results in using artificial intelligence in various fields. It has been suggested that artificial intelligence could be used to detect the presence or absence of the queen bee, but few studies have explored this before. This thesis examines bee sounds using machine learning algorithms to determine whether a queen bee is present. Features were extracted from hive sound recordings using Fast Fourier Transform and Mel-Frequency Cepstral Coefficients methods. Subsequently, 18 different attribute combinations were created for these methods, considering their statistical significance and cases where 95% and 99% of the variance was explained by principal component analysis. These attributes were used as input to eight different classifier algorithms, which have been commonly used in the literature. The highest classifier performance was achieved with Mel-Frequency Cepstral Coefficients with Principal Component Analysis by explaining 99% of the variance, and the Decision Tree algorithm, achieving 92.59% accuracy. The results demonstrate that algorithms can be implemented on affordable electronic cards by achiving similar performance with the state-of-the-art studies.