Machine Learning challenges

Focus on the Machine Learning course challenges solved by the students of the Artificial Intelligence and Machine Learning course at ‘Politecnico Di Bari’. The aim of these challenges is to test both the theoretical knowledge and coding skills about the Machine Learning topics covered in the lessons. The students had 2 hours and half for each challenge to implement a solution using Machine Learning principles and algorithms.

The topics covered by these challenges are:
Gradient descent algorithm for univariate linear regression, multivariate linear regression, polynomial regression, logistic regression.
Data preprocessing: Min-max and Z-score normalization, feature selection, removing outliers. Cross validation: hold-out method, K-folds method, random subsampling.
Model evaluation: learning curves, metrics, ROC curve.
Neural networks and Support Vector Machines.
Unsupervised learning techniques: K-Means, K-Medoids, GMM, Hierarchical clustering, DBSCAN.
Dimensionality reduction: PCA and Kernel PCA.

Each challenge is briefly described below:

• BOOK PRICE PREDICTION: exploit data about sale books to train a Machine Learning model capable of predicting the price of a book, starting from the available information about it.
• IMAGES CLUSTERING AND CLASSIFICATION: classify 28x28 grayscale images. Apply dimensionality reduction using PCA, divide the dataset in 10 subsets using clustering algorithms, computer the ideal number of clusters using Elbow method and compute Silhouette Coefficient applied to each clustering method used.
• LOGISTIC REGRESSION: predict an object unknown feature, based on other features and a training dataset. Apply min-max and/or Z-score normalization, build a multiclass logistic regression model with L2 regularization using full-batch Gradient Descent.
• NEURAL NETWORK LOGISTIC REGRESSION: predict an object unknown feature, based on other features and a training dataset using a Neural Network coded by scratch. Apply Z-score normalization.

If you publish any work which uses the code stored in this project, please cite the following creators:
Sergio Abascià, Gianluca Azzollini, Alberto Carlo Maria Mancino

Developers
Sergio Abascià
Gianluca Azzollini
Alberto Carlo Maria Mancino

Contacts
We are happy to help you with any question. Please contact us on our mails:
[email protected]
[email protected]
[email protected]