Intel_Image_Classification_Challenge_Convolutional_Neural_Network

Built and used my own convolutional neural network and 4 other pre-trained CNNs (Resnet-50, VGG-16, Inception-V3, Xception) on the dataset of images provided by Intel. Trained these models on a dataset containing images of buildings, forests, streets, mountains, glaciers and sea to find out which is the best convolutional neural network model, to predict correctly for a new image. Trained the models on around 14,000 images and tested them on 3000 images. Used pre-trained models, as they have already been trained on thousands of images and optimized them by tuning hyperparameters, training for longer time, adding more layers and performing data augmentation to achive best results.

According to the validation accuracy, the best performing models on the above dataset were:

1. Xception (92%)

2. Inception_V3 (91.3%)

3. Vgg-16 (88.73%)

4. Self-Built Model (83.53%)

5. ResNet-50 (69.4%)