Machine Learning Project

by Sabrina Fonseca Pereira and Maria Sousa

About the project

The purpose of this project is to implement and use different machine learning models to classify glass fragments into 6 categories, based on 9 different forensic measurements.

A neural network and a decision tree were implemented from scratch and compared to the corresponding scikit-learn implementation. Other models provided by the scikit-learn library were also used as a way to explore their efficacy in solving this classification problem.

The repository

code

• implementations.py contains the implementation of our machine learning models from scratch
• project.ipynb the notebook with model predictions and analysis.

data

• df_train.csv training set
• df_test.csv test set

dt-visualization

• decision-tree.json was made from the dictionary outputted by the decision tree training
• tree.svg is tree visualisation generated from the json file (generated with https://vanya.jp.net/vtree/)

report.pdf the report with project findings and conclusions.

The jupyter notebook contains the function calls, visualisations and tests with the sklearn library. The notebook is dependent on files in the code folder.