Analysis of a multiclass classification problem by Lasso Logistic Regression and Singular Value Decomposition to identify sound patterns in queenless bee colonies

Abstract

This study presents an analysis of a multiclass classification problem to identify queenless states by monitoring bee sound in two possible cases; a strong and healthy colony that lost its queen and a reduced population queenless colony. The sound patterns were compared with patterns of healthy queenright colonies. Five colonies of Carniola honey bee were monitored by using a system based on a Raspberry Pi 2 and omnidirectional microphones placed inside the hives. Feature extraction was carried out by Mel Frequency Cepstral Coefficients (MFCCs) method. A multiclass model with three outcome variables was constructed. For feature selection and regularization, a Lasso logistic Regression model was used along with one vs all strategy. To provide visual evidence and examine the results, data was analyzed by scatter plots of Singular Value Decomposition (SVD). The results show that is possible to detect the queenless state in both cases. Queenless or healthy colonies can generate slightly different patterns and the data clusters of the same condition tend to be close. The proposed methodology can be applied for the analysis of more conditions in bee colonies.