Does Noise Affect Housing Prices? A Case Study in the Urban Area of Thessaloniki

Introduction

This repository consists of the coding infrastructure required to run the experiments mentioned in the corresponding paper. Additionally, the datasets used are provided via GDrive links further below.

Installation Requirements

First, make sure you have installed all the needed packages. To do so, just run:

pip install .

Important note: If you are using a Windows machine, you will have to manually download both GDAL and rasterio. Visit the two websites and download the versions that match your Python version. Then run the following:

pip install -U pip
pip install /path/to/gdal/wheel/*.whl
pip install /path/to/rasterio/wheel/*.whl

MLflow

The entire machine learning development is tracked by MLflow. The parameters and the metrics of each experiment are logged in the mlruns folder and can be viewed in via the MLflow UI. To do so, run:

mlflow ui

Noise

In the noise folder there are primitive data and scripts needed to reconstruct noise data for the municipalities of Thessaloniki, Neapoli and Kalamaria, in the urban area of Thessaloniki. The primitive data are basically high resolution heatmaps with noise information. These heatmaps were enriched with latitude and longitude details through the process of Georeferencing. So, for each heatmap there is the corresponding *.tif file, which spatial information.

The reconstructed noise datasets are located in the noise/generated-data folder.

File	Description
thessaloniki_day_noise.parquet.gzip	Day noise dataset for Thessaloniki and Neapoli (overall noise)
thessaloniki_night_noise.parquet.gzip	Night noise dataset for Thessaloniki and Neapoli (overall noise)
kalamaria_day_noise.parquet.gzip	Day noise dataset for Kalamaria (overall noise)
kalamaria_night_noise.parquet.gzip	Night noise dataset for Kalamaria (overall noise)
kalamaria_aviation_day_noise.parquet.gzip	Aviation day noise dataset for Kalamaria
kalamaria_aviation_night_noise.parquet.gzip	Aviation night noise dataset for Kalamaria

All files, both generated and primitive data, can be found in this GDrive folder.

Openhouse Properties

The final version of the Openhouse properties must be moved from GDrive's properties folder to the data folder of this repository. There are 2 version each with a different radius (50 and 100 meters).

Experiments

Each model has its own experiment file which can be executed via command line by providing the required arguments. These are:

• --noise {noise_id} 0 -> no noise, 1 -> both day and night noise, 2 -> average of day and night noise, 3 -> only day noise, 4 -> only night noise
• --mode {hyperparameter_tuning_id} 'grid_search' -> Grid Search, 'bayesian' -> Baysian, else -> Random
• --area {area_id} 'A' -> center, 'B' -> Triandria, Toumpa, Harilaou, 'C' -> Kalamaria
• --experiment-name {name} The name of the experiment to be created
• --experiment-tags {tags} A dictionary containing any extra information
• --experiment-id {id} The id of an already existing experiment

Examples

To train an XGBoost with Bayesian optimization in Kalamaria using both day and night noises run:

py .\decision_trees_experiment.py --noise 1 --mode "bayesian" --area "C"  --experiment-name "Test"

Evaluation

The implementation of the evaluation techniques can be found in the evaluate.py file.