Scripts and Methodological Workflow

Reproducible analytical pipeline for Urban PM2.5 research

Purpose of this section

This page documents the analytical scripts that implement the methodological pipeline developed in the doctoral thesis and extended in this project.

Rather than focusing on algorithmic performance, the scripts reflect a pipeline-oriented research strategy, where coherence, transparency, and reproducibility are prioritised over model benchmarking.

All scripts are openly available in the project repository and can be executed independently.

Methodological position

The analytical workflow is structured around the following principles:

  • Sequential and traceable execution.
  • Explicit handling of temporal structure.
  • Separation between data preparation, exploration, modelling, and evaluation.
  • Responsible integration of emerging computational paradigms.
  • Full reproducibility using open tools and documented scripts.

This approach is aligned with the methodological framework defended in the doctoral thesis and presented here as an example of research continuity.

Script overview

01 — Data preparation and preprocessing

01_data_and_preprocessing.R

This script implements the first stage of the pipeline: transforming raw observational data into an analysis-ready dataset.

Main objectives - Load raw daily PM2.5 data. - Parse and validate temporal information. - Identify and handle missing values. - Ensure internal consistency and traceability.

Output - A clean dataset stored in the data/ directory, used by all subsequent scripts.

This step establishes the analytical baseline of the entire workflow.

02 — Exploratory and correlation analysis

02_exploratory_and_correlation_analysis.R

This script focuses on understanding the structural properties of the PM2.5 time series.

Main objectives - Descriptive statistical analysis. - Daily time series visualisation. - Monthly and seasonal variability assessment. - Detection of missing temporal segments. - Exploratory correlation patterns.

Outputs - High-resolution figures stored in figures_tiff/. - Console summaries for transparency and traceability.

The goal is not prediction, but contextual understanding of the data-generating process.

03 — Classical modelling and evaluation

03_modelling_and_evaluation.R

This script implements simple, interpretable classical models to establish a methodological reference.

Main objectives - Construction of a persistence baseline (lag-1). - Linear regression with temporal structure. - Time-based train–test splitting. - Evaluation using MAE and RMSE. - Visual comparison between observed and modelled values.

Models are intentionally kept simple to emphasise methodological validation rather than optimisation.

04 — Quantum exploratory analysis (demonstrative)

04_quantum_exploratory_analysis_qiskit.py

This script demonstrates the extensibility of the pipeline to emerging quantum computing architectures.

Key characteristics - Implemented in Python due to the native Qiskit ecosystem. - Uses a quantum kernel-based regression model. - Executed under explicit NISQ constraints. - Training performed on a deliberately reduced subset. - No comparison with classical models is intended.

Interpretation Reported metrics serve solely as a sanity check, confirming that the quantum block integrates correctly into the existing analytical workflow.

No claims of quantum advantage or scalability are made.

Execution order

Scripts are designed to be executed sequentially:

  1. Data preparation
  2. Exploratory analysis
  3. Classical modelling
  4. Quantum methodological extension

This structure ensures full traceability and reproducibility.

Reproducibility and openness

All scripts: - Are fully documented. - Rely exclusively on open-source tools. - Produce results programmatically. - Can be executed independently by third parties.

The technical documentation for execution details is provided in the corresponding README.md files within the repository.

Final remark

This script-based architecture is presented as an example of methodological maturity in applied data science research, where analytical coherence and transparency take precedence over technological novelty.

It is intended to support both academic research and educational use. .