You can access the project by clicking on this link.

Asteroid Classification Using SVM and Kernel SVMs

In collaboration with Mathieu Graf, this machine learning project aimed to classify asteroids as hazardous or non-hazardous by employing Support Vector Machines (SVM) and kernel SVMs. This initiative was a deep dive into SVM analysis, enhancing our understanding of kernel functions and their impact on classification tasks.

Project Overview

Leveraging a dataset from NASA’s Center for Near Earth Object Studies (CNEOS), we scrutinized 4687 observations, each detailing the physical and orbital properties of asteroids. Our goal was to create a predictive model to determine the potential hazard of an asteroid based on these characteristics.

Methodological Approach

Our approach began with analyzing the data linearity using a linear SVM, followed by applying non-linear SVMs. Through extensive cross-validation, we optimized the model parameters, focusing on the number of support vectors, the objective function’s value, and the accuracy of classifications. Data visualization techniques, such as principal component analysis, were pivotal in interpreting the outcomes.

Innovative Explorations

A significant portion of our research was dedicated to experimenting with different kernel functions, including linear, polynomial, and RBF kernels. The Laplacian kernel was particularly notable for its robustness against outliers and noise, which was evident in our model with a gamma value of 1.

Key Learnings

The project accentuated the criticality of choosing the right kernel functions and the delicate balance needed in parameter tuning. The performance metrics provided insights into the capabilities and limitations of SVMs and kernel SVMs in classification tasks.

Working with Mathieu Graf on this project was a demonstration of our combined technical expertise and collaborative synergy. It was an invaluable experience that refined our analytical skills and enhanced our data science proficiency for future endeavors.