🚀 Time Series Distance Estimation

Irregular Time Series Processing and Regression Pipeline (NASA ADS / arXiv:2311.04470)

This repository contains the published paper and implementation notebooks for a large-scale time series modeling project available on NASA ADS (arXiv:2311.04470).

The project implements a complete end-to-end analytical pipeline for extracting structured features from irregular observational time series and estimating physical distances using regression modeling.

Author: Kevin Mota da Costa

Portfolio: https://costakevinn.github.io

LinkedIn: https://linkedin.com/in/costakevinnn

---

🎯 Project Purpose

The objective of this project was to design a robust time series and regression workflow capable of extracting meaningful patterns from large, noisy, irregularly sampled datasets.

The pipeline integrates:

• Irregular time series processing
• Period detection using Lomb–Scargle (Astropy)
• Phase-folding and feature engineering
• Supervised regression modeling
• Quality filtering and outlier handling
• Validation against external reference datasets (OGLE-IV)

Although the application is astrophysical, the workflow directly translates to financial time series, sensor monitoring, industrial analytics, and forecasting systems.

---

🧠 System Architecture

The analytical workflow follows a structured lifecycle:

Raw irregular time series → Period detection (Lomb–Scargle) → Frequency transformation → Phase-folding → Feature extraction → Regression modeling → Distance estimation → Benchmark validation

This structure mirrors production-grade time series engineering pipelines.

---

📊 Dataset

• 4,700+ real observational samples
• OGLE-IV reference catalog
• Irregular sampling
• Heteroscedastic measurement noise

The project includes statistical validation against benchmark datasets to ensure predictive reliability.

---

🔬 Methods & Techniques

• Irregular time series analysis
• Lomb–Scargle periodogram (Astropy)
• Phase-folding of periodic signals
• Feature engineering from temporal structure
• Statistical regression (SciPy curve_fit)
• Kernel Density Estimation
• Outlier filtering and quality control
• Benchmark validation

---

📈 Visual Results

Time Series → Frequency Transformation

(Lomb–Scargle applied to irregular data)

---

Feature Engineering from Periodic Data

(Phase-folded time series)

---

Regression Model on 4,700+ Samples

(Period–luminosity relationship)

---

Model Validation vs Reference Dataset

(Comparison with OGLE-IV benchmark)

---

Probability Density Estimation

---

Feature-to-Target Mapping

---

📂 Repository Contents

• docs/paper.pdf Published scientific paper (Portuguese)

• notebooks/time_series_distance_pipeline.ipynb Complete pipeline: loading, period detection, feature engineering, regression, validation

• notebooks/equation_43_derivation.ipynb Supporting statistical derivations and modeling details

---

🔗 Publication

NASA ADS / arXiv:2311.04470 https://ui.adsabs.harvard.edu/abs/arXiv:2311.04470

Note: The original publication links to an older GitHub account. This repository hosts the updated implementation and paper under my current profile.

---

🛠 Tech Stack

Programming

Python

Data & Scientific Computing

• NumPy
• Pandas
• SciPy
• Astropy

Time Series & Statistics

• Lomb–Scargle periodogram
• Kernel Density Estimation
• Statistical regression
• Outlier filtering

Visualization

• Matplotlib

---

🔬 Capabilities Demonstrated

• Large-scale irregular time series processing
• Feature engineering from periodic signals
• Regression modeling on noisy data
• Data validation against benchmark datasets
• Analytical reproducibility
• Structured ML-style pipeline design

---

🌐 Portfolio

This project is part of my Machine Learning portfolio: 👉 https://costakevinn.github.io

---

License

MIT License — see LICENSE for details.