A Hybrid Deep Transfer Learning Approach For The Detection Of Vector-Borne Diseases

Abstract

Vector-borne diseases considerably impact the worldwide population’s health and economic well-being. However, training deep-learning models requires significant time and training data. Therefore, a unique hybrid transfer learning approach was proposed for detecting vector-borne diseases (VBD) to solve these issues while retaining high accuracy. In the first phase, malaria and Lyme benchmark datasets were obtained. Then VGG16, VGG19, MobileNetV2, and DenseNet 169 were compared to the hybrid model results (MobileNetV2+DenseNet 169). The effectiveness of the hybrid transfer learning method was evaluated using several performance measures, namely precision, loss, accuracy, AUC and RMSE. On the malaria dataset, the proposed model (MobileNetV2+DenseNet 169) achieved the most excellent classification accuracy of 99.9%, and on the Lyme dataset, 99.3%.