PREDICTION OF POTENTIALLY HAZARDOUS ASTEROIDS

The problem statement is to develop an ML model to classify hazardous asteroids. It helps to easily identify and track asteroids which are hazardous and allow timely risk management. The Approach aims at outlining subgroups of near earth asteroids (NEAs) with high PHA concentration in them.

OVERVIEW OF THE WORKFLOW:

1.Collect the data from relevant sources

2.Do the Data Preprocesssing

3.Find the PHAs from the real data

4.Make more data using virtual asteroids

5.Choose the ML Model which will give maximum accuracy

VIRTUAL ASTEROIDS

We synthetically increase amount of data by generating virtual asteroids and computing MOID for them. We do this inorder to unlock more deeper insights about our data for asteroids.

Uniform Dataset : We generate new dataset assuming the various oarameters follow uniform distribution. We do this so that we are able to create a uniform space in ‘w’ and ‘q’ so that clustering algorithm can work efficiently.

Non-Uniform Dataset : These are distributions that approximate distributions of real NEAs.

Clustering using DBSCAN

1.Ability to Find Arbitrary-Shaped Clusters: DBSCAN can detect clusters of various shapes, making it suitable for datasets with non-uniformly shaped clusters, unlike some other algorithms like K-Means, which assume spherical clusters.

2.Robust to Noise: It's capable of distinguishing noisy data points as outliers or noise, ensuring that they don't disrupt the cluster formation. This makes it more robust in the presence of data imperfections.

3. No Need to Specify the Number of Clusters: Unlike K-Means, which requires you to specify the number of clusters beforehand, DBSCAN automatically identifies the number of clusters within the data.

4. Efficiency: DBSCAN efficiently handles large datasets and is less sensitive to the order of data points, which can be an issue with hierarchical clustering.

Results

• We used SVM, Random Forest and XG Boost model for classification in all the domains and below are the best results

Domain 1

• Highest accuracy of classification of hazardous and non hazardous asteroid in Atiras and Atens is achieved using SVM Linear Kernel i.e 91.34%
• Highest accuracy of classification of hazardous and non hazardous asteroid in Apollo is achieved using SVM RBF Kernel i.e 86.85%
• Highest accuracy of classification of hazardous and non hazardous asteroid in Amors is achieved using SVM RBF Kernel i.e 98.31%

Domain 2

• Highest accuracy of classification of hazardous and non hazardous asteroid in Atiras and Atens is using XG Boost Model i.e 85.91%
• Highest accuracy of classification of hazardous and non hazardous asteroid in Apollo is using XG Boost Model i.e 94.61%

Domain 3

• Highest accuracy of classification of hazardous and non hazardous asteroid in Apollo group is achieved using XG Boost Model i.e 94.69%

REFERENCES

1.Prediction of Orbital Parameters for Undiscovered Potentially Hazardous Asteroids by Vadym Pasko

2.Hazardous Asteroid Classification through Various Machine Learning Techniques by Anish Si

3.https://github.com/nomad-vagabond/asterion?search=1