Datasets:

Machine-Earning
/

CISIR-datasets

license: mit
task_categories:
  - time-series-forecasting
tags:
  - regression
  - tabular-data
  - imbalanced-learning
  - sep-forecasting
  - solar-physics

Highly Imbalanced Regression with Tabular Data in SEP and Other Applications

This repository contains the datasets used in the paper Highly Imbalanced Regression with Tabular Data in SEP and Other Applications.

Code: https://github.com/Machine-Earning/CISIR

Abstract

We investigate imbalanced regression with tabular data that have an imbalance ratio larger than 1,000 ("highly imbalanced"). Accurately estimating the target values of rare instances is important in applications such as forecasting the intensity of rare harmful Solar Energetic Particle (SEP) events. For regression, the MSE loss does not consider the correlation between predicted and actual values. Typical inverse importance functions allow only convex functions. Uniform sampling might yield mini-batches that do not have rare instances. We propose CISIR that incorporates correlation, Monotonically Decreasing Involution (MDI) importance, and stratified sampling. Based on five datasets, our experimental results indicate that CISIR can achieve lower error and higher correlation than some recent methods. Also, adding our correlation component to other recent methods can improve their performance. Lastly, MDI importance can outperform other importance functions.

Overview

This dataset collection is part of the official implementation of CISIR (Correlation, Involution, Stratified Importance Regression), a novel method for highly imbalanced regression with tabular data. The method is particularly applicable to SEP (Solar Energetic Particle) forecasting research for NASA and other domains requiring accurate prediction of rare, high-impact events.

Key Contributions (from the associated paper)

Correlation-aware loss function that considers the correlation between predicted and actual values
Monotonically Decreasing Involution (MDI) importance weighting that outperforms traditional convex importance functions
Stratified sampling strategy that ensures rare instances are included in mini-batches
Comprehensive evaluation on five highly imbalanced datasets with imbalance ratios > 1,000

Datasets

We evaluate our method on five highly imbalanced datasets:

SEP Datasets

SEP-EC: Forecasts the change (delta) in proton intensity based on features from electron intensity and CMEs (Coronal Mass Ejections).
SEP-C: Forecasts peak proton intensity based on CME characteristics.

Other Datasets

SARCOS: Estimates the torque vector based on joint-state inputs for a 7-DOF robot arm.
Blog Feedback (BF): Forecasts the number of comments based on textual, temporal, and engagement features
Online News Popularity (ONP): Estimates the number of shares of an article based on content, topic, and sentiment attributes

All datasets exhibit high imbalance ratios (ρ > 1,000), making them ideal for evaluating highly imbalanced regression methods.

Data Availability

You can download the combined dataset archive here: https://huggingface.co/datasets/Machine-Earning/CISIR-datasets/resolve/main/CISIR-data.zip

Citation

If you find this dataset useful in your research, please consider citing the associated paper:

@inproceedings{moukpe2024cisir,
  title={Highly Imbalanced Regression with Tabular Data in SEP and Other Applications},
  author={Moukpe, Josias K. and Chan, Philip K. and Zhang, Ming},
  booktitle={Proceedings of the IEEE International Conference on Machine Learning and Applications (ICMLA)},
  year={2024},
  organization={IEEE}
}