Prediction of different physiological conditions of riverine buffaloes (bubalus bubalis) based on their vocal cues through machine learning algorithms and a conventional statistical model.

To understand the requirements of animals their calls can be analysed. This potentially enables specific and more precise individual care under different emotional and physiological conditions. This study was conducted to identify three different conditions (oestrus, delayed milking and isolation) of buffaloes based on vocalization patterns. A total of 600 acoustic samples of buffaloes for each condition were collected under different conditions consisting of 300 records for confirming and 300 for non-confirming of a particular condition. Important acoustic features like amplitude (P), total energy (P²s), pitch (Hz), intensity (dB), formants (Hz), number of pulses, number of periods, mean period (sec) and unvoiced frames (%) were extracted using the MFCC (mel frequency cepstrum coefficients) technique. Algorithms (model) were trained by partitioning the acoustic data into training and validation sets to develop predictive models. Three different ratios were assessed: 60%-40%, 70%-30% and 80%-20%. Decision tree models were optimized based on decision and average square error (probability) options and other parameters were set to default values of the software package to deveop the best model. The performance of algorithms was evaluated on the parameter accuracy rate. Decision tree models predicted the physiological conditions oestrus, isolation and delayed milking with an accuracy of 66.1, 84.3 and 71.3%, respectively, while the logistic regression model predicted with an accuracy rate of 59.5, 71.1 and 65.7%, respectively, and the artificial neural network (ANN) model predicted these three conditions with 77.7, 85.2 and 79.4% accuracy, respectively. The ANN model was found to be best on the basis of minimum misclassification rate (on 80%-20% portioning). However, decision tree algorithms also provided the additional information that intensity (maximum), amplitude (minimum) and formant (F1) are the most important features of vocal signals to identify physiological conditions like oestrus, isolation and delayed milking respectively in dairy buffalo.