In this repo you will find my work for the Machine Learning exam at UniMi which aims to create a good machine learning model to binary classify images of Chihuahuas and Muffins from this dataset. In particular, you will find both the code - with cut outputs to ensure readability - and the report in which I have widely explained the methodology and the steps taken to reach the final result. Still, the report was not the one presented at the exam. Indeed, I integrated it with many theoretical concepts useful to understand how the convolutional neural networks work.
The objective of this project is pursued through the implementation of Convolutional Neural Networks (CNNs) thanks to the Keras API of Tensorflow. After many tests and hyperparameter tuning, the final model was able to correctly classify around 9 pictures out of 10 as demonstrated by the 5-Fold Cross-Validation.
To guarantee reproducibility of results, I made available the final model in .h5 format, feel free to download the file and to load it with Keras.
I am open to receive any feedback from you, enjoy! 💻