GSoC 2025 aeon Introduction: Somto Onyekwelu - Applying for Project #2 (ML Forecasting Evaluation)

[Somto Onyekwelu]

Hello aeon Mentors and Community,

Following the guidance on the aeon GSoC page encouraging communication, I'm writing to briefly introduce myself as a prospective GSoC 2025 contributor. My name is Somto Onyekwelu, a Computer Science undergraduate from Nigeria, and I am very excited to be applying for Project #2: Forecasting - Implementing and evaluating machine learning forecasters.

Predicting future trends is fascinating, and the opportunity to implement and rigorously evaluate non-deep learning ML forecasters like SETAR-Tree within a leading time series library like aeon aligns perfectly with my interests in practical ML and robust evaluation.

My full proposal (submitted via the GSoC portal) details a plan to:

• Implement SETAR-Tree based on [Godahewa et al., 2023].
• Develop reusable time series feature engineering utilities to improve framework transparency for ML models (addressing a stated project goal).
• Rigorously evaluate SETAR-Tree against key aeon baselines (e.g., Dummy, KNeighborsTimeSeries, Rocket, TimeSeriesForest) using standard metrics (MAE/RMSE/MASE).
• Uniquely: Analyze the robustness of these forecasters against controlled noise in historical data (additive noise, missing values) – providing insights into real-world reliability.
• (Stretch Goal): Provide a baseline wrapper for a LightGBM forecaster.

To demonstrate my commitment and readiness, I've created a Proof-of-Concept repository showing foundational skills in time series handling, lagging, and basic ML forecasting workflow with Python/Pandas/Scikit-learn:

https://github.com/SomtoOnyekwelu/gsoc-2025-aeon-ml-forecast-evaluation

Regarding the evaluation in Project #2, I was wondering: besides standard accuracy metrics, how much emphasis is typically placed on computational efficiency benchmarks when comparing new forecasting algorithms within aeon's framework?

Thank you for providing this platform for interaction. I am very enthusiastic about the potential to contribute to aeon and learn from the community this summer!

Best regards,
Somto Onyekwelu
Project PoC: https://github.com/SomtoOnyekwelu/gsoc-2025-aeon-ml-forecast-evaluation

[Somto Onyekwelu]

Hi all,

Just wanted to quickly confirm that my final proposal for Project #2 has been successfully submitted via the official GSoC portal before the deadline.

Excited about the possibility and looking forward to receiving follow up and hearing the outcome in May!

Best regards,
Somto Onyekwelu

[Somto Onyekwelu]

Hi all,

Just wanted to quickly confirm that my final proposal for Project #2 has been successfully submitted via the official GSoC portal before the deadline. 

The title of my project proposal is:

Evaluating and Enhancing Machine Learning Forecasters: Implementation, Benchmarking, and Robustness Analysis

by Somto Onyekwelu

Excited about the possibility and looking forward to hearing the outcome in May!

Best regards,
Somto Onyekwelu

[Somto Onyekwelu]

Good day aeon,

I have submitted my first pull request, which documents a verified exception in KNeighborsTimeSeriesClassifier's initialization and aims to improve user-facing API clarity. This is part of my efforts to familiarise myself with aeon's codebase and guidelines.

I'm very excited to keep learning and contributing, and I'm grateful to be part of this community!

Best regards,

Somtochukwu Onyekwelu