부산대학교 도서관

Human-Elephant conflict has become very common in the forest borders causing phenomenal increase in death of elephants as well as loss of human life. Elephants face a shortage of resources such as food and water, resulting in Human–Elephant Conflict. The presence of elephants can be predicted using elephant voice detection. We propose a machine learning based system for detecting elephant voice and predicting the presence of elephants in the forest border areas. Existing methods for elephant voice detection in literature require large feature data dimensions. In the proposed system, feature extraction methods combined with Principal Component Analysis (PCA) that greatly reduces feature data dimension is proposed to improve the performance metrics of the recorded elephant voice samples. A Support vector machine (SVM) classifier is used for the predictive model in this work. The proposed system is validated using the cross validation method and the performance metrics such as Accuracy, Sensitivity, Specificity, Precision, F1 Score and Computation time are evaluated. It is observed that with the proposed approach the average accuracy is 93.32% and feature data dimension is 1422 compared to an average accuracy of 83.5% obtained and feature data dimension of 18,882 with methods using Mel-Frequency Cepstral Co-efficient (MFCC)