Research in Machine Learning for Biodiversity MeasurementObjective:Conduct research to explore the utility of machine learning combined with other techniques to accurately categorize audio spectrograms in an effort to measure biodiversity in Sonoma County, California.Techniques:Utilized Matlab to employ support vector machines (SVM) and experiment with various kernel functions (linear, polynomial, radial basis function) to optimize classification performance.Conducted thorough result analysis using diverse visualization techniques such as confusion matrices to assess classification accuracy, ROC curves for evaluating the trade-off between sensitivity and specificity, and precision-recall curves to understand the performance in terms of true positive rates.Leveraged Matlab and PyTorch for building, training, and evaluating convolutional neural networks (CNNs). This involved designing custom CNN architectures and also applying transfer learning techniques on established models like AlexNet and ResNet to improve classification accuracy.Achieved a high identification rate (over 90%) by meticulously testing various activation functions (ReLU, Sigmoid, Tanh) and network structures (number of layers, filter sizes, etc.).Data Analysis and Results:Conducted extensive data preprocessing to handle noise and ensure the quality of input data.Performed hyperparameter tuning using grid search and cross-validation to find the optimal parameters for the models.Applied feature selection techniques to identify the most significant features contributing to accurate classification, thereby improving model efficiency and performance.Analyzed the impact of different training strategies and regularization techniques on model performance to prevent overfitting and enhance generalization.

Data Analysis, Technical Support, Customer Service, Travel