AegPupaCNN: A Custom Convolutional Neural Network and Comparative Models for Sex Classification of Aedes aegypti Pupae Using Individual Images

Data Description Dataset Title: Data from: AegPupaCNN: A Custom Convolutional Neural Network and Comparative Models for Sex Classification of Aedes aegypti Pupae Using Individual Images Description of Experimental Efforts: This dataset was generated to evaluate the efficacy of Convolutional Neural Networks (CNNs) for automating the sex classification of Aedes aegypti pupae, a process essential for the Sterile Insect Technique (SIT). Experimental data collection involved rearing Ae. aegypti mosquitoes in a controlled laboratory setting across six generations. Individual pupae were imaged using a digital microscope (SVBONY M-201) from both lateral and ventral views to capture sexually dimorphic morphological features. The dataset provided here contains the test performance metrics and raw prediction outputs for five CNN architectures (AegPupaCNN, ResNet50, MobileNet, VGG16, and InceptionV3), which were trained on five generations of training data and cross-validated/ tested on an independent sixth generation. Geographic Location: Quezon City, Philippines. Funding: Department of Science and Technology – Science Education Institute (DOST-SEI), Philippines. Files and variables File: AegPupaCNN_lateral.xlsx Description: Contains raw prediction data and calculated performance metrics for the custom AegPupaCNN model evaluated on lateral view images of Aedes aegypti pupae. Sheet 1 provides instance-level predictions for the independent test set. Sheet 2 provides summary performance metrics. Variables: Sheet 1 (Raw Data): Filename: Unique identifier string for the specific image file tested (e.g., "female\F1..."). True Label: The actual biological sex of the pupa (Ground truth). Values: "female", "male". Predicted Label: The classification output generated by the model. Values: "female", "male". Sheet 2 (Metrics): Metric: Name of the statistical performance measure. Values: "Accuracy", "Precision", "Recall", "F1-Score", "ROC-AUC". Value: The calculated numerical score for the corresponding metric (Range: 0.0–1.0 or 0–100%). File: InceptionV3_lateral.xlsx Description: Contains raw prediction data and calculated performance metrics for the InceptionV3 model evaluated on lateral view images of Aedes aegypti pupae. Sheet 1 provides instance-level predictions for the independent test set. Sheet 2 provides summary performance metrics. Variables Same variables as AegPupaCNN_lateral.xlsx. File: MobileNet_lateral.xlsx Description: Contains raw prediction data and calculated performance metrics for the MobileNet model evaluated on lateral view images of Aedes aegypti pupae. Sheet 1 provides instance-level predictions for the independent test set. Sheet 2 provides summary performance metrics. Variables Same variables as AegPupaCNN_lateral.xlsx. File: ResNet50_lateral.xlsx Description: Contains raw prediction data and calculated performance metrics for the ResNet50 model evaluated on lateral view images of Aedes aegypti pupae. Sheet 1 provides instance-level predictions for the independent test set. Sheet 2 provides summary performance metrics. Variables Same variables as AegPupaCNN_lateral.xlsx. File: VGG16_lateral.xlsx Description: Contains raw prediction data and calculated performance metrics for the VGG16 model evaluated on lateral view images of Aedes aegypti pupae. Sheet 1 provides instance-level predictions for the independent test set. Sheet 2 provides summary performance metrics. Variables Same variables as AegPupaCNN_lateral.xlsx. File: AegPupaCNN_ventral.xlsx Description: Contains raw prediction data and calculated performance metrics for the custom AegPupaCNN model evaluated on ventral view images of Aedes aegypti pupae. Sheet 1 provides instance-level predictions for the independent test set. Sheet 2 provides summary performance metrics. Variables Same variables as AegPupaCNN_lateral.xlsx. File: InceptionV3_ventral.xlsx Description: Contains raw prediction data and calculated performance metrics for the InceptionV3 model evaluated on ventral view images of Aedes aegypti pupae. Sheet 1 provides instance-level predictions for the independent test set. Sheet 2 provides summary performance metrics. Variables Same variables as AegPupaCNN_lateral.xlsx. File: MobileNet_ventral.xlsx Description: Contains raw prediction data and calculated performance metrics for the MobileNet model evaluated on ventral view images of Aedes aegypti pupae. Sheet 1 provides instance-level predictions for the independent test set. Sheet 2 provides summary performance metrics. Variables Same variables as AegPupaCNN_lateral.xlsx. File: ResNet50_ventral.xlsx Description: Contains raw prediction data and calculated performance metrics for the ResNet50 model evaluated on ventral view images of Aedes aegypti pupae. Sheet 1 provides instance-level predictions for the independent test set. Sheet 2 provides summary performance metrics. Variables Same variables as AegPupaCNN_lateral.xlsx. File: VGG16_ventral.xlsx Description: Contains raw prediction data and calculated performance metrics for the VGG16 model evaluated on ventral view images of Aedes aegypti pupae. Sheet 1 provides instance-level predictions for the independent test set. Sheet 2 provides summary performance metrics. Variables Same variables as AegPupaCNN_lateral.xlsx. Code/software Software needed to view the data: The data files are in standard Microsoft Excel (.xlsx) format. They can be viewed and edited using any spreadsheet software. Software used to generate the data: The data was generated using custom scripts written in Python 3.12. The following key libraries were used for image preprocessing, model training, and performance evaluation: TensorFlow (Keras API): For building and training the CNN architectures. Scikit-learn: For calculating performance metrics (Precision, Recall, F1-Score, ROC-AUC) and generating classification reports. OpenCV (cv2): For image preprocessing tasks including background removal and resizing. Pandas: For structuring the output data and exporting to .xlsx format. Access information Other publicly accessible locations of the data: None. Data was derived from the following sources: The data provided in this repository are original and were generated by the authors during the study. They were not derived from external public datasets. The biological material (Aedes aegypti mosquitoes) used to generate the image data was obtained from the mosquito-rearing facility at the Philippine Nuclear Research Institute.