Electron Energy Flux Prediction using Deep Learning

Project Report

https://sugatagh.github.io/dsml/projects/electron-energy-flux-prediction/

Kaggle Notebooks

Part $1$. Exploratory Data Analysis: https://www.kaggle.com/code/sugataghosh/electron-energy-flux-prediction-part-1-eda

Part $2$. Modeling: https://www.kaggle.com/code/sugataghosh/electron-energy-flux-prediction-part-2-modeling

Overview

• In McGranaghan et al. (2021), the authors have considered the problem of modeling electron particle precipitation from the magnetosphere to the ionosphere. They attempted to address it through a new database, using machine learning tools to extract useful information from it.

• Based on that database, we aim to predict electron total energy flux, which is a continuous variable.

• A detailed exploratory data analysis on the dataset is carried out. In particular, we observe that there are clear groups among the feature variables. We investigate the group-wise correlation structure through averaging the pairwise correlation coefficients.

• We use the insights obtained from EDA in the data preprocessing stages, which consists of feature extraction, data transformation, feature scaling, and principal component analysis.

• We build a neural network and tune it to predict electron total energy flux.

• The final model obtains a root mean square error (RMSE) of $1.531904$, a mean absolute error (MAE) of $1.075939$, and a coefficient of determination $(R^2)$ of $0.699466$ on the test set.