Using Cats vs Dog DB for Classification by using CNN and optimizing accuracy with different method. In this project, I was able to achieve a classifier accuracy of 94.64% by making my own model.

By building an appropriate neural network (with convolutional and max pooling layers) suitable for image classification, and using techniques to optimize the training process: data normalization Improving the CNN with BatchNormalization layer and Dropouts ass needed when there is an indication of overfitting Data Augmentation to artificially expand the amount of data Testing hyper parameters and finding the ones which yield the best results Using the regularization ReduceLROnPlateau

The method: monitoring and checking the accuracy and the cost functions of the validation and the test sets, we want to manipulation the prooccess using the tools above, the graphs behave: Making them more and more consolidated and until they reach the lowest possible point where we reach a good amunnot of loss I'll mantion that for that kind of leaeeningg the graphes are more noisy by nature, a stochistic grdiant diesnt is very noisy we prefer to use the adam ooptimazer for less noise and faster Convergence due to the adaptivee Convergence and the momentum coomponnent

Also trying to make the grapths with less spikes foor - to get more accurate result, so smothing it is an option. a good modle will look like this:

ה