DecisionTreeCoursework

This coursework implements a decision tree algorithm identify rooms based on WiFi signal strengths collected from a mobile phone. Trees are trained, validated and tested on two datasets: wifi_db/clean_dataset.txt and wifi_db/noisy_dataset.txt.

The coursework can be split into 4 major steps:

1. Loading Data.
2. Creating Decision Trees.
3. Evaluation of trees using 10-fold cross validation.
4. Pruning and evaluation of trees using nested 10-fold cross validation..

An evaluation of pruned and unpruned trees, and a visualization of a tree trained on the entire clean dataset is done in the report found in docs/Decision_Tree_Coursework_Report.pdf.

Setting up Virtual Environment

A virtual environment should be setup before installing packages in prerequisites. Python virtual environment can be activated via:

cd your_project_directory
python3 -m venv venv
source venv/bin/activate

Prerequisites

The packages in requirements.txt should be installed by running:

pip install --upgrade pip
pip install -r requirements.txt
python3 -c "import numpy as np; import torch; print(np); print(torch)"

Running the Program

Run src/main.py from the root directory to

• Create pruned and unpruned decision trees for each of the clean and noisy dataset.
• Print out evaluation metrics for each tree.
• Saves a visualisation of a tree trained on the clean dataset in images.

The program can be ran with the following command:

python3 src/main.py

Output

After running the program, the following data are being printed for trees trained on wifi_db/clean_dataset.txt and wifi_db/noisy_dataset.txt:

1. Average Tree Depth
2. Average of Confusion Matrices over 10 folds
3. Average Overall Accuracy
4. Precision
5. Recall
6. F1-Score