Queenless honeybee acoustic patterns
Description
The data set consists of acoustic patterns of five Carniola honeybee colonies located in Zacatecas, Mexico. The colonies were monitored from March to May 2018. The beehives analyzed were two healthy queenright colonies with a huge population, two healthy colonies with moderate populations, and a queenless colony with a low population. Two queens were removed from two colonies to create the queenless condition. The acoustic pattern of healthy colonies is presented in Folder 1, and the experiment of queens removed in Folder 2. The hypothesis is that the queenless state of a beehive can be identified by comparing the acoustic pattern with the pattern of a healthy colony by using machine learning techniques and feature extraction methodologies. The data is presented with no previous preprocessing step, in the analysis Mel frequency cepstral coefficients were used for feature extraction, and the classification was carried out by using support vector machines, neural networks, and logistic regression, all of them with high classification rate. The data can be analyzed by using singular value decomposition or principal component analysis to understand the cluster, and how they change when the queens were removed from the colonies.
Files
Steps to reproduce
The acoustic samples were acquired by using a monitor system based on a Raspberrry Pi 2 and electret microphones (Max4466), and a dspic microcontroller was used as ADC converter. The samples were recorded with a sampling frequency of 4 kHz at 12 bits resolution, every sample is 30s long, per day 144 samples were recorded at 10 minutes intervals. The reproduction of the samples can be done by using the function 'sound' of MATLAB. For appropriate analysis, the offset of the signal must be removed.