Machine Learning

This repository encompasses the projects developed for the Machine Learning course, which aimed to familiarize students with different Machine Learning models and data processing techniques.

Team:

• Alexandre Sobreira
• André Dias
• Tiago Rodrigues

Professors:

• André Falcão
• Andreia Teixeira
• Nuno Garcia

First Home Assignment - Decision Trees and Linear Regression

This home assignment focused on the utilization of basic machine learning models for classification and regression tasks and their full evaluation through the use of a test set and an independent validation set.

Models used:

Regression: Linear Regression, Regularized Linear models (Ridge and Lasso regularization) and Decision Tree Regressors

Classification: Logistic Regression and Decision Tree Classifiers

Data:

The data corresponds to the Parkinsons Telemonitoring Data Set, which is composed of a range of biomedical voice measurements from 42 people with early-stage Parkinson's disease recruited to a six-month trial of a telemonitoring device for remote symptom progression monitoring.

Assignment Objectives:

• Producing the best regression model for 'motor UPDRS'
• Producing the best binary classification model assuming as positive all instances with values of total_UPRDS> 40 and as negatives all remaining cases

Second Home Assignment - Data pre-processing and more complex classification models

This home assignment focused on pre-processing of the given data, through imputation and feature selection, the utilization of varied classification models (including ensemble models) and the optimization of their parameters.

Models used for classification:

• Decision Trees
• Logistic Regression
• Naïve Bayes
• K-Nearest Neighbors
• Support Vector Machines
• Bagging
• Random Forests
• AdaBoost
• Gradient Boost
• Extreme Gradient Boost

Data:

The data was provided by the Professors and was adaptaded from the QSAR biodegradation Data Set, which focuses on the study of the relationships between chemical structure and biodegradation of molecules.

Assignment Objectives:

• Data pre-processing (normalization, imputation, initial feature selection)
• In depth feature selection using correlation metrics and machine learning models (Decision Trees, Stepwise methods, Random Forests)
• Producing the best Classification model (parameter optimization included)

Final Project - Time series prediction

This project focused on data pre-processing of a given dataset and the development of accurate prediction models.

Models used for regression:

• Linear Regression
• Random Forests
• Extreme Gradient Boost

Data:

The data corresponds to the "World Bank Data" from Kaggle, which contains 3 datasets related to country population, fertility rate and life expectancy from 1960 up to 2016 for several countries.

Project Objectives:

• Data pre-processing (feature encoding, creating a time series)
• Add additional information to make better models
• Make predictions for 2017 and 2018 and compare to the real results obtained from the World Bank Data