Swiss Residential Perceived Livability Assessment

Overview

This project implements a Fuzzy Inference System to assess the perceived livability of Swiss residential dwellings using the Computing with Words framework. The system transforms quantitative environmental simulations into linguistic assessments that align with human perception.

Research Questions

1. RQ1: How can a fuzzy-logic-based web application represent urban residential livability in a way that is understandable and meaningful for citizens?
2. RQ2: How do potential users perceive the usability and perceived usefulness of the proposed livability tool?

Project Structure

swiss-livability-assessment/
├── README.md                # This file
├── pyproject.toml           # Project configuration
├── .python-version          # Python version
│
├── src/                     # Core fuzzy inference modules
│   ├── __init__.py
│   ├── membership_functions.py  # Fuzzy membership functions
│   ├── rule_base.py         # Fuzzy inference rules
│   ├── fuzzy_system.py      # Mamdani FIS implementation
│   ├── feature_alignment.py # Feature alignment layer (raw → FIS)
│   └── assessment.py        # Centralized assessment thresholds and labels
│
├── app/                     # Web application
│   ├── __init__.py
│   ├── web_app.py           # Flask web server
│   └── templates/           # HTML templates
│       ├── base.html
│       ├── index.html
│       ├── explore.html
│       ├── assess.html
│       └── filter.html
│
├── scripts/                 # Utility scripts
│   ├── common.py            # Shared utilities (paths, data loading, constants)
│   ├── prepare_full_features.py  # Data preparation (two-stage)
│   ├── validate_alignment.py     # Alignment validation
│   └── run_assessment.py    # Run FIS assessment + visualizations
│
├── data/
│   ├── raw/                 # Original dataset
│   │   └── swiss-dwellings-v3.0.0/
│   └── processed/           # Processed data
│       ├── dwellings_full.csv      # Aligned features
│       └── feature_alignment.json  # Alignment config
│
├── results/                 # Output results
│   ├── figures/             # Visualizations
│   └── outputs/             # CSV and reports
│
└── docs/                    # Documentation
    ├── literature_review.md
    └── web_app_guide.md

Installation

Prerequisites

Install uv (recommended Python package manager):

# macOS/Linux
curl -LsSf https://astral.sh/uv/install.sh | sh

# Windows
powershell -ExecutionPolicy ByPass -c "irm https://astral.sh/uv/install.ps1 | iex"

Install dependencies

uv sync

Usage

Data Preprocessing

Prepare full features from raw data (two-stage processing):

uv run python scripts/prepare_full_features.py

This script performs:

1. Stage 1: Extract raw features from Swiss Dwellings dataset (original units)
2. Stage 2: Fit and apply feature alignment to match FIS input universes

Validate the feature alignment:

uv run python scripts/validate_alignment.py

Run Assessment

Run the full FIS assessment with visualizations:

uv run python scripts/run_assessment.py

Run assessment only (skip visualization generation):

uv run python scripts/run_assessment.py --no-viz

Run Web Application

uv run python -m app.web_app

Then open http://localhost:5001 in your browser.

Features:

• Home (/): Overview of the livability assessment system
• Explore (/explore): Browse dwelling statistics and sample data
• Assess (/assess): Manually input dwelling parameters to compute FLI score
• Filter (/filter): Filter and sort all dwellings by livability criteria
• i18n Support: Available in English, German (Deutsch), and French (Français)

Dataset

Swiss Dwellings v3.0.0 (3,171 processed dwellings)

Raw features (from simulations.csv and locations.csv):

• Noise: window_noise_traffic/train_day/night (dBA)
• Daylight: sun_*_mean at noon for 3 seasons (klx)
• Views: view_sky_p80, view_greenery_p80 (steradians)
• Location: walkshed_* POI counts (202 categories)

Feature Alignment

The Feature Alignment layer transforms raw dataset values to match the FIS input universes, ensuring all dimensions properly contribute to the fuzzy inference.

FIS Design

The alignment uses a simplified approach where membership function universes are calibrated directly to match the raw data distributions, eliminating complex scaling transformations.

Reference data statistics (Swiss Dwellings v3.0.0):

• view_sky: 0–0.13 sr (median ~0.029, 95th pctl ~0.05)
• view_greenery: 0–0.06 sr (median ~0.01, 95th pctl ~0.026)
• daylight: 0–3.91 klx (noon illuminance)
• location_poi: 3–2662 counts → log10: ~0.6–3.43

Transformations

Variable	Transformation	Purpose
Noise	Pass-through, ≤0 → missing	Already in dBA
Daylight	Pass-through (klx), cap at 6.0	MF universe calibrated to klx
View sky	Pass-through (sr), cap at 0.13	MF universe matches raw data range
View greenery	Pass-through (sr), cap at 0.06	MF universe matches raw data range
Location POI	log10(count+1), cap at 3.5	Compress long-tail distribution

Configuration

Alignment parameters are saved to data/processed/feature_alignment.json:

{
  "daylight_klx_cap": 6.0,
  "view_sky_max": 0.13,
  "view_greenery_max": 0.06,
  "location_poi_log_max": 3.5
}

These values define the upper bounds for each FIS input universe, ensuring consistency between batch processing and the web API.

FLI Results

After feature alignment, the Fuzzy Livability Index distribution (n=2,988 valid dwellings):

Statistic	Value
Mean	37.57
Std Dev	15.36
Min	13.17
Median	40.00
Max	77.11

Linguistic Label Distribution:

Label	Count	Percentage
Excellent	114	3.8%
Good	683	22.9%
Fair	1,362	45.6%
Poor	829	27.7%

Feature Correlations with FLI:

• noise_lden: −0.74 (strong negative — more noise = lower livability)
• noise_lnight: −0.73 (strong negative)
• daylight: +0.35 (moderate positive)
• view_sky: +0.35 (moderate positive)
• view_greenery: +0.20 (moderate positive)
• location_poi: −0.16 (weak negative)

Standards Applied

WHO 2018 Environmental Noise Guidelines

• Road traffic: Lden < 53 dB, Lnight < 45 dB
• Railway: Lden < 54 dB, Lnight < 44 dB
• Aircraft: Lden < 45 dB, Lnight < 40 dB

EN 17037 Daylight Provision

• Minimum: 300 lux over 50% of area, 100 lux over 95% of area
• Medium: 500 lux over 50% of area, 300 lux over 95% of area
• High: 750 lux over 50% of area, 500 lux over 95% of area

Key References

1. Mendel, J. M. (2002). An architecture for making judgments using computing with words. International Journal of Applied Mathematics and Computer Science, 12(3), 325-335.
2. WHO (2018). Environmental Noise Guidelines for the European Region.
3. EN 17037:2018. Daylight in buildings.
4. Zadeh, L. A. (1999). From computing with numbers to computing with words. IEEE Transactions on Circuits and Systems, 45(1), 105-119.

Authors

• Hao Wang
• Emmanuel Cazzato