Comparative Analysis of Machine Learning Models for Hazardous
Asteroid Classification
Nandika P S
1
, Lakshana S
1
, Shruti Lakshmi V
1
and Manju Venugopalan
2
1
Department of Electrical and Electronics Engineering Amrita School of Engineering,
Bengaluru Amrita Vishwa Vidyapeetham, India
2
Department of Computer Science and Engineering Amrita School of Computing,
Bengaluru Amrita Vishwa Vidyapeetham, India
Keywords:
Hazardous Asteroids, Planetary Defense, Machine Learning, Classification, AdaBoost, Decision Tree, CAT
Boost, Feature Selection, Orbital Parameters, Absolute Magnitude
Abstract:
Automating the classification of hazardous asteroids is crucial for planetary defense, enabling timely and ac-
curate threat assessment. This study delves into leveraging machine learning models to classify hazardous
asteroids comprehensively. Utilizing a dataset rich in features like absolute magnitude and orbital parame-
ters, the research navigates through preprocessing, feature selection, model selection, and evaluation stages.
Through meticulous preprocessing, the dataset is refined to ensure optimal performance in subsequent mod-
elling endeavors. Sophisticated feature selection techniques identify the most discriminative features crucial
for accurate asteroid classification. Various algorithms, including Decision Tree, AdaBoost, and CAT Boost,
are evaluated across different metrics. AdaBoost and Random Forest emerges as a standout performer, demon-
strating superior performance with an F1 score of 0.87 and 0.88 AdaBoost achieves an accuracy rate of 95
respectively, highlighting its robustness and potential as a formidable tool in planetary defense. These findings
underscore the critical role of precise asteroid classification in mitigating potential threats to planetary safety
by enabling proactive measures against identified hazardous asteroids.
1 INTRODUCTION
The threat posed by hazardous asteroids to Earth has
long been a subject of concern, prompting extensive
research efforts aimed at developing accurate classi-
fication methods to safeguard our planet. With the
potential to cause catastrophic damage upon impact,
the identification and characterization of hazardous
asteroids are crucial for devising effective mitigation
strategies and ensuring the safety of humanity. Tra-
ditional methods of asteroid classification often rely
on spectral analysis and orbital observations, which,
while effective, can be time-consuming and labor-
intensive. However, recent advancements in machine
learning present a promising avenue for automating
and enhancing this process. Machine learning al-
gorithms have demonstrated remarkable capabilities
across various domains (Aravind et al., 2019; Kumar
et al., 2024; Madhusoodanan et al., 2024), from image
recognition (Sowmya and Deepika, 2014),(Neena and
Geetha, 2018) to natural language processing (Venu-
gopalan and Gupta, 2022). In the context of aster-
oid classification, these algorithms offer the poten-
tial to streamline and improve existing classification
methods significantly. Machine learning algorithms
can efficiently analyze vast data and identify com-
plex patterns, accelerating classification and provid-
ing more accurate predictions about the hazardous-
ness of newly discovered asteroids. The paper inves-
tigates the effectiveness of various machine learning
models in classifying hazardous asteroids, leverag-
ing a dataset consisting of relevant features extracted
from astronomical observations. By training models
on labelled datasets containing information about the
characteristics of asteroids and their hazardousness,
machine learning algorithms can learn to identify sub-
tle patterns and make accurate pre- dictions about the
hazardousness of asteroids. This research aims to
enhance our understanding of asteroid classification
methodologies and contribute to planetary defense ef-
forts. We aim the accurately recognize the best ma-
chine learning models for classifying hazardous as-
teroid. The study also seeks to understand the factors
influencing model performance and optimize them for
642
P S, N., S, L., V, S. L. and Venugopalan, M.
Comparative Analysis of Machine Learning Models for Hazardous Asteroid Classification.
DOI: 10.5220/0013599400004664
Paper published under CC license (CC BY-NC-ND 4.0)
In Proceedings of the 3rd International Conference on Futuristic Technology (INCOFT 2025) - Volume 2, pages 642-648
ISBN: 978-989-758-763-4
Proceedings Copyright © 2025 by SCITEPRESS – Science and Technology Publications, Lda.

real-world applications. Furthermore, this research
has implications beyond asteroid classification, as the
methodologies and techniques developed can be ap-
plied to other domains facing similar classification
challenges. Through machine learning, we enhance
our ability to identify and mitigate potential threats
from asteroids and other celestial objects. Ultimately,
our goal is to contribute to the collective efforts aimed
at safeguarding our planet and ensuring the continued
safety and well-being of future generations.
2 MANUSCRIPT PREPARATION
Previous research has extensively explored various
machine learning techniques for asteroid classifica-
tion, reflecting the growing interest in leveraging
computational methods to address planetary defense
challenges. (J. Smith, 2020) conducted a compre-
hensive study utilizing decision trees and neural net-
works decision trees and neural networks to clas-
sify asteroids based on orbital parameters, provid-
ing valuable insights into the applicability of these
methods in asteroid classification tasks. This study
explores efficacy of different machine learning algo-
rithms in handling complex astronomical data and
demonstrated promising results in accurately identi-
fying hazardous asteroids. Building upon this founda-
tion, (M. Brown, 2018) (delved into the application of
support vector machines (SVM) and random forests
for asteroid classification using spectral data. Their
work contributed significantly to the understanding
of machine learning-based asteroid classification by
feature extraction and selection in improving accu-
racy of the classification. By leveraging spectral in-
formation, their approach demonstrated the potential
of machine learning models to discern subtle differ-
ences in asteroid compositions and classify them ac-
cordingly. In a similar vein, (R. Jones, 2019) the
study explored ensemble learning methods like bag-
ging and boosting for asteroid classification, aiming
to enhance predictive performance and robustness by
combining multiple classifiers. Their research show-
cased the effectiveness of ensemble techniques in
enhancing classification accuracy and demonstrated
their utility in handling uncertainties and noise in as-
teroid data. Expanding the scope of feature selec-
tion techniques,(L. Zhang, 2021) proposed a novel ap-
proach based on genetic algorithms for asteroid clas-
sification. Genetic algorithms mimic the process of
natural selection to iteratively evolve a set of features
that maximize classification performance. Their work
demonstrated the efficacy of evolutionary approaches
in identifying relevant features and reducing dimen-
sionality, thereby improving the efficiency and inter-
pretability of classification models. Recent studies
by (S. Kim, 2022) explored deep learning approaches
for asteroid classification, leveraging convolutional
neural networks (CNNs) to extract features from as-
teroid images. By harnessing the power of CNNs,
their research achieved remarkable classification ac-
curacy and illustrated the capability of deep learning,
in handling complex high- dimensional data. Simi-
larly, (Y. Wu, 2021) investigated the use of transfer
learning techniques for asteroid classification, lever-
aging knowledge from pre-trained models to enhance
classification performance. Their work underscored
the importance of leveraging existing knowledge and
resources to address challenges in asteroid classifi-
cation effectively. Further pushing the boundaries
of classification accuracy, (X. Li, 2023) introduced
a novel hybrid model combining deep learning and
evolutionary algorithms for asteroid classification. By
integrating the strengths of both approaches, their re-
search achieved state-of-the-art performance in terms
of accuracy and computational efficiency. Addition-
ally, the supervised approach by(A. Johnson, 2022),
proposed a supervised classification for analyzing
and detecting potentially hazardous asteroids, adding
to the repertoire of classification techniques tailored
specifically for identifying hazardous asteroids.
3 DATASET DESCRIPTION
The dataset used in this study is sourced from Nasa1.
The dataset comprises 34 features, including absolute
magnitude, estimated diameter (in kilometers and
miles), relative velocity, and orbit period, among
others. The target variable categorizes asteroids as
either hazardous or non-hazardous.
Nasa1:href{https://www.kaggle.com/datasets/
lovishbansal123/nasa-asteroids-classification}.
4 PROPOSED METHODOLOGY
The following pipeline in Fig. 1 shows the structural
process for data preprocessing, data handling, mod-
elling and evaluating in a machine learning workflow.
The process begins with data visualization, using rep-
resentations such as, pie chart, histogram and box plot
are used to understand the dataset. This is followed by
data preprocessing, which involves handling missing
values, standardizing the binary data to 0’s and 1’s and
feature selection, to select the best and relevant fea-
tures using the K-select method. The preprocessing
step also involves balancing the dataset using SMOTE
Comparative Analysis of Machine Learning Models for Hazardous Asteroid Classification
643

(Synthetic Minority Over-sampling Technique). Once
the data have been pre-processed, model selection us-
ing various models such as, SVM, naive bayes, KNN,
random forest, and others, is conducted. Following
the model selection, the dataset is split into training
and testing. After this, evaluation of the dataset is
done using performance metrics like, accuracy, preci-
sion recall and F1 score. Results of these models are
compared using ROC curve and F1 score. Best model
is selected on the basis of these evaluation metrics.
The methodology is explained in detail in the follow-
ing sub sections.
4.1 Data Visualization
Data visualization is the first process that is done to
explore and understand the dataset. It tells whether
the dataset is balanced or imbalanced, gives an anal-
ysis of the selected features in the dataset and also
gives information about the outliers present in the
dataset. Graphical representations like pie chart, his-
togram and box plot are used.
4.1.1 Visualizing class distribution
Fig. 2 represents the class distribution of the target
class in the dataset hazardous or non-hazardous in the
data. As observed in Fig. 2, 83.9%belong to non-
hazardous, while 16.1% belong to hazardous, a typi-
cal indication that it is an imbalanced dataset. The be-
low dataset consists of a feature named” Hazardous”.
This feature classifies the asteroid as hazardous or
non-hazardous. Hence, this is a binary feature with
true indicating hazardous asteroid and false indicating
non-hazardous asteroid. Absolute Magnitude feature
represents the brightness of the asteroid, so, it comes
under hazardous, estimated diameter represents the
diameter of the asteroid, which is also classified as
hazardous. The relative velocity shows the velocity of
the asteroid, and the fastness relative to earth, which,
is also classified as hazardous. Miss distance talks
about the distance of the asteroid from earth which
can also be classified as hazardous.
4.1.2 Visualizing Data Distribution of Input
Feature For Categorical and Numerical
Attributes:
Fig. 3 represents a box plot which shows the distri-
bution of various variables of features in the dataset.
The positive skewness of the box plot indicates that
there are few larger and farther dataset, creating out-
liers in the distribution. The wide range values in few
features as shown in Fig. 3 shows that it is a highly
variable dataset. Fig. 4 shows the distribution of
Figure 1: ML methodology pipeline for asteroid
classification
INCOFT 2025 - International Conference on Futuristic Technology
644

Figure 2: Class distribution of the dataset
Figure 3: Box plot to visualize outliers
the data for a few features. Right skewed indicates
that most of the variables in the dataset are smaller
slower, while some being faster and farther. Fig. 4
also highlights that common events like low diame-
ters are common compared to rare events like large
diameters. The wide range as shown in figure indi-
cates the variability in the dataset.
4.2 Data Preprocessing
In the phase of data preprocessing, the dataset is effec-
tively handled using various techniques to ensure its
integrity and completeness. Methods such as impu-
tation or deletion are utilized to address missing val-
ues, striking a balance between data preservation and
maintaining accuracy. Data preprocess- ing, involves
handling missing values, standardizing the data, fea-
ture selection, to select the best and relevant fea-
tures using the K-select method. The preprocessing
step also involves balancing the dataset using SMOTE
(Synthetic Minority Over- sampling Technique). The
preprocessing is explained in detail in the following
sub sections.
Figure 4: Histogram Visualization for continuous features
4.2.1 Imputation
Imputation is the process of handling any missing val-
ues in the dataset. It is handled by replacing a missing
in the dataset. It is an important process as ML algo-
rithms cannot handle missing values.
4.2.2 Standardizing Data
Standardizing data involves scaling the numeric val-
ues of the dataset between 0 and 1. In this case, the
binary data present in the target column is converted
to 0’s and 1’s.
4.2.3 Data Balancing
Given that the dataset was highly imbalanced, a cru-
cial step in the preprocess is to address this im-
balance through data balancing techniques. Specif-
ically, the Synthetic Minority Over-sampling Tech-
nique (SMOTE) (K. Li and Fang, 2014; D. Dablain
and Chawla, 2023; G. A. Pradipta and Ismail, 2021)
is applied to balance the classes, which is shown in
fig.5.
4.2.4 Feature Selection
Feature selection significantly enhances effectiveness
of machine learning models asteroid classificati- Fig.
5. Class Distribution of the balanced dataset after
SMOTE on. The selected features provide valuable
insights into the underlying patterns and characteris-
Comparative Analysis of Machine Learning Models for Hazardous Asteroid Classification
645

Figure 5: Class Distribution of the balanced dataset after
SMOTE
tics of asteroids, allowing for more accurate classifi-
cation outcomes.
4.3 Model Selection
Model selection is used identify suitable models for
binary classification problem of asteroid classifica-
tion. In this study, various algorithms including De-
cision Tree, AdaBoost, Cat Boost, Random Forest,
Bagging, SVM, KNN, and Naive Bayes are consid-
ered for model evaluation. Each model is assessed
based on its ability to accurately classify asteroids as
hazardous or non-hazardous using performance met-
rics such as accuracy, F1 score, and recall. By com-
paring the performance of different models, insights
can be gained on the strength and weakness of these
algorithms, which facilitates the decision regarding
the selection of the most appropriate model.
4.4 Training and Testing
Training and testing are important steps in the devel-
opment of the model, where the dataset is dividing
into 2 subsets,70% of the data is taken for training,
while 30% of the data for testing. During training,
the model gets trained to identify various relations
and patterns using the training dataset, which is later
tested using the test dataset.
4.5 Model Evaluation
Evaluation of models involves a detailed analysis us-
ing performance metrics, such as accuracy, preci-
sion, and recall and F1 score. These metrices give
an idea on the model’s capability to make accurate
predictions and to handle different classes within the
dataset. To understand the impact of this balancing
on model performance, the F1 scores of the dataset
were plotted. By calculating the performance metrics
for both before and after the feature selection process,
the model performance was evaluated. This visualiza-
tion helped in assessing how data balancing and fea-
ture selection affected the model’s performance and
its capability in handling minority class. This com-
parative analysis provided insights into how selecting
the most relevant features influenced overall perfor-
mance of the model and its ability in making correct
predictions.
5 RESULTS AND DISCUSSION
The experiment initiated on the imbalance data-based
feature selection was applied, which extracted 11 fea-
tures out of 34 features. Results of the experimented
classifiers with and without feature selection on the
original dataset is given in Fig. 6. And Fig.7.
The dataset was subjected to oversampling us-
ing SMOTE. The results of experimented classifiers
with and without feature selection display in Fig. 8.,
and Fig. 9. By evaluating various machine learning
models highlight significant variations in their perfor-
mances. Fig. 6, Fig. 7 shows a comparative analysis
of various models using bar graph. The F1 score was
calculated for various models for both before and af-
ter balancing as shown in Fig. 6 and Fig. 7. These
illustrate the F1 score for entire dataset and feature
selected dataset using bar graph. From Fig. 6 it is
observed that the F1 score for the entire dataset be-
fore balancing is significantly larger compared to the
F1 score of the feature-selected dataset. However,
Fig. 7. shows a reduced difference in F1 score val-
ues between train and test dataset. This indicates the
improved model performance and reduced overfitting.
Fig. 8 shows that shows the Receiver Operating Char-
acteristics (ROC) before and after feature selection for
random forest and Fig.9. shows the ROC curve before
and after feature selection for adaboost. These ROC
curves showcase the performance of best 2 algorithms
having highest F1 score value: Random Forest and
Adaboost. We can see from the figures that the true
positive rate of the entire dataset is comparatively less
than that of the feature selected dataset. This visual
comparison tells the effectiveness of feature selection
in enhancing the predictive power of these algorithms.
A gamut of machine learning models is then scru-
tinized, spanning Decision Trees, AdaBoost, CAT
boost, Random Forest, Bagging, SVM, KNN, and
Naive Bayes classifiers shown in Table 1. Leverag-
ing performance metrics like accuracy, ROC curves,
INCOFT 2025 - International Conference on Futuristic Technology
646

Figure 6: F1 score across classifiers on test data before data
balancing
Figure 7: F1 score across classifiers on test data after data
balancing.
Figure 8: ROC curve for Random Forest and Adaboost be-
fore feature selection
Figure 9: ROC curve for Random Forest and Adaboost after
feature selection
Table 1: Model Performance Comparison
Model Accuracy Precision Recall F1 Score
AdaBoost 0.89 0.85 0.85 0.87
Cat Boost 0.87 0.86 0.86 0.87
Random Forest 0.88 0.87 0.87 0.88
Bagging (RF, CatBoost) 0.88 0.86 0.86 0.88
SVM 0.85 0.81 0.80 0.85
KNN 0.85 0.83 0.83 0.83
Naive Bayes 0.87 0.80 0.77 0.76
and confusion matrices, this rigorous evaluation en-
ables structured framework and empirical evidence to
drive advancements in the field of planetary defense.
Through its meticulous methodology and insightful
analyses, the code embodies the potential of machine
learn into address complex challenges at the intersec-
tion of astronomy and artificial intelligence. Model
evaluation after comparison is shown in Table II.
Table 2: Model evaluation after Comparison
Model Accuracy Precision Recall F1 Score
AdaBoost 1.00 1.00 1.00 1.00
Cat Boost 0.99 0.99 0.99 0.99
Random Forest 1.00 1.00 1.00 1.00
Bagging
(random forest, 1.00 1.00 1.00 1.00
catboost)
SVM 0.94 0.94 0.94 0.94
KNN 0.89 0.88 0.88 0.89
Naive Bayes 0.93 0.94 0.93 0.93
6 CONCLUSION
In conclusion, by meticulously navigating each phase
of the machine learning pipeline, from initial data
exploration to final model assessment, the proposed
methodology exemplifies a systematic and compre-
hensive approach to addressing the intricate chal-
lenges of asteroid classification. The processing of
data, selection of features, and visualization plays a
vital role in setting the effective model training, en-
suring the integrity of the dataset, reducing dimen-
sionality, and uncovers crucial insights into the under-
lying patterns and relationships within the data. This
groundwork provides researchers with a deeper un-
derstanding of the nuances inherent in asteroid data.
The subsequent evaluation of a diverse range of ma-
chine learning models offers a nuanced perspective on
their performance and suitability for asteroid classifi-
cation tasks. From Decision Trees to Random Forests,
each algorithm undergoes rigorous scrutiny, with per-
formance metrics offering quantitative assessments of
accuracy and model robustness. Among the ensemble
of models examined, Random Forest emerges as the
standout performer, demonstrating superior accuracy
and predictive capabilities on the test set. This finding
underscores the effectiveness of ensemble methods in
Comparative Analysis of Machine Learning Models for Hazardous Asteroid Classification
647

tackling complex classification tasks and underscores
the importance of leveraging a variety of algorithms
to optimize model performance. Moreover, the com-
prehensive evaluation of models not only identifies
the most effective classifier but also provides informa-
tion into the strength and weakness of the algorithms.
This knowledge helps researchers to select the most
appropriate model for asteroid classification, thereby
informing future endeavours in planetary defense and
space exploration.
REFERENCES
A. Johnson, e. a. (2022). Supervised classification for anal-
ysis and detection of potentially hazardous asteroid.
IEEE Transactions on Aerospace and Electronic Sys-
tems, 68(4):198–213.
Aravind, T., Reddy, B. S., Avinash, S., and G., J. (2019). A
comparative study on machine learning algorithms for
predicting the placement information of under gradu-
ate students. IEEE I-SMAC (IoT in Social, Mobile,
Analytics and Cloud), pages 542–546.
D. Dablain, B. K. and Chawla, N. V. (2023). Deepsmote:
Fusing deep learning and smote for imbalanced data.
IEEE Transactions on Neural Networks and Learning
Systems, 34(9):6390–6404.
G. A. Pradipta, R. Wardoyo, A. M. I. N. H. S. and Ismail,
M. (2021). Smote for handling imbalanced data prob-
lem: A review. In Proceedings of the Sixth Inter-
national Conference on Informatics and Computing
(ICIC), pages 1–8, Jakarta, Indonesia. SCITEPRESS.
J. Smith, e. a. (2020). Asteroid classification using deci-
sion trees and neural networks. Astronomy Journal,
123(4):567–578.
K. Li, W. Zhang, Q. L. and Fang, X. (2014). An improved
smote imbalanced data classification method based on
support degree. In Proceedings of the International
Conference on Identification, Information and Knowl-
edge in the Internet of Things, pages 34–38, Beijing,
China. SCITEPRESS.
Kumar, R. H. et al. (2024). Rainfall prediction enhance-
ment using smote and machine learning algorithms.
In 2024 15th International Conference on Computing
Communication and Networking Technologies (ICC-
CNT). IEEE.
L. Zhang, e. a. (2021). Feature selection for asteroid classi-
fication using genetic algorithms. IEEE Transactions
on Geoscience and Remote Sensing, 89(2):789–802.
M. Brown, e. a. (2018). Machine learning techniques for
asteroid classification based on spectral data. Journal
of Planetary Sciences, 45(2):210–225.
Madhusoodanan, D. et al. (2024). A comprehensive com-
parison of machine learning techniques for heart dis-
ease prediction. 2024 15th International Conference
on Computing Communication and Networking Tech-
nologies (ICCCNT).
Neena, A. and Geetha, M. (2018). Image classification us-
ing an ensemble-based deep cnn. In Advances in In-
telligent Systems and Computing, volume 709, pages
445–456. SCITEPRESS.
R. Jones, e. a. (2019). Ensemble learning methods for as-
teroid classification. IEEE Transactions on Aerospace
and Electronic Systems, 65(3):1345–1358.
S. Kim, e. a. (2022). Deep learning approaches for asteroid
classification. IEEE Transactions on Pattern Analysis
and Machine Intelligence, 44(5):1123–1137.
Sowmya, D. S. K. P. and Deepika, J. (2014). Image classi-
fication using convolutional neural networks. Interna-
tional Journal of Scientific & Engineering Research,
5(6):06/2014.
Venugopalan, M. and Gupta, D. (2022). A reinforced active
learning approach for optimal sampling in aspect term
extraction for sentiment analysis. Expert Systems with
Applications, 209:118228.
X. Li, e. a. (2023). Hybrid deep learning and evolutionary
algorithms for asteroid classification. IEEE Transac-
tions on Evolutionary Computation, 28(1):245–259.
Y. Wu, e. a. (2021). Transfer learning techniques for aster-
oid classification. IEEE Transactions on Cybernetics,
77(3):678–692.
INCOFT 2025 - International Conference on Futuristic Technology
648