Model Purpose: Detection of tuberculosis infection or disease.
Data Source: Tuberculosis (TB) Chest X-ray Database
Features Used: The input features are the chest X-ray images. These images are preprocessed and augmented using techniques like rescaling, rotation, shifting, shearing, zooming, and horizontal flipping.
Model Architecture: The base architecture used is DenseNet121, a convolutional neural network (CNN) pre-trained on ImageNet. A custom dense layer is added on top of the base model for binary classification.
Dataset Size: There are 2940 images in the dataset.
Training Accuracy: The model achieved a training accuracy of 98%.
Batch Size: The batch size used for training and validation data generators is set to 32.
Number of Models: The code trains an ensemble of 3 models with different architectures or initializations.
Epochs and Early Stopping: Each model is trained for 10 epochs, with early stopping applied based on validation loss to prevent overfitting.
Model Compilation: The models are compiled using the Adam optimizer with a learning rate of 0.0001 and binary cross-entropy loss function. Accuracy is used as the evaluation metric.