Quantum Machine Learning for Exoplanet Transit Detection

Abstract

This project investigates the use of hybrid quantum–classical machine learning models for detecting exoplanet transits in stellar light curves. Synthetic transit signals are generated to create a large training dataset, and models are evaluated on real observations from the Kepler space telescope. Classical models (Logistic Regression, Random Forest, SVM) are compared with a Quantum Neural Network (QNN). Results show that classical ensemble models remain competitive while QNN performance is comparable but not superior on small datasets. The real data source is the Kepler Mission via the Mikulski Archive for Space Telescopes (MAST).

Research Question

Can hybrid quantum neural networks improve exoplanet transit detection compared to classical machine learning models when trained on synthetic data and evaluated on real Kepler observations?

Methodology

Synthetic Data Generation ↓ Kepler Light Curve Download ↓ Feature Extraction ↓ PCA Dimensionality Reduction ↓ Model Training ↓ Evaluation on Real Kepler Stars

Dataset Description

Synthetic Dataset

• 5000 generated light curves
• 50% transit / 50% non-transit

Real Dataset (Kepler Mission)

• 92 Kepler stars
• 44 transit hosts
• 48 non-transit stars

Results

Model Performance (Real Kepler Evaluation)

Model	Accuracy	ROC-AUC
Logistic Regression	0.505	0.50
Random Forest	0.546	0.548
SVM	0.495	0.50
Quantum Neural Network	0.485	0.517

Qubit Scaling Experiment

Qubits	Accuracy
2	0.546
3	0.495
4	0.526

Conclusion: Increasing qubits did not consistently improve accuracy.

Visual Results

• ROC curve: results/plots/roc_curve.png
• Confusion matrix: results/plots/confusion_matrix.png
• Model comparison: results/plots/model_comparison.png
• QNN scaling plot: results/plots/qnn_qubit_scaling.png

Reproducing the Experiment

1. Install dependencies

pip install -r requirements.txt

2. Generate synthetic dataset

python -m src.synthetic_transit --n_samples 5000 --negative_fraction 0.5 --out data/raw/synthetic_transits.csv

3. Download Kepler light curves

python -m src.download_kepler --targets Kepler-10 Kepler-22 Kepler-69

4. Train models

python -m src.train --config config/config.yaml

Future Work

• Training on a larger Kepler dataset
• Deeper quantum circuits
• Alternative quantum feature encodings
• Evaluation on TESS (Transiting Exoplanet Survey Satellite) light curves