A Comparative Analysis of Data Resampling Techniques for Addressing Class Imbalance in Mosquito Classification

Mosquitoes, as vectors of life-threatening diseases such as dengue, Zika, and West Nile virus, present a significant public health challenge. Precise and rapid identification of mosquito species is crucial for early detection and prevention of these diseases. Deep learning techniques have proven to be highly effective for automatic visual insect classification tasks. However, the imbalanced distribution of mosquito species within training datasets severely impacts the algorithm’s classification performance. This study investigates data sampling solutions for learning from imbalanced mosquito images dataset. We transfer-learn InceptionV3, a pre-trained convolutional neural network to classify images of two vector mosquito species Aedes Aegypti and Anopheles stephensi. The effectiveness of these data sampling solutions is evaluated and compared in terms of their impact on the performance of CNN classifier.