A Comparative Analysis of XGBoost Model and AdaBoost Regressor
for Prediction of Used Car Price
S. Naveen Kumar Reddy
*
and S. Magesh Kumar
Department of Computer Science Engineering Saveetha School of Engineering,
Saveetha Institute of Medical and Technical Sciences, Saveetha University, Chennai, Tamil Nadu, 602105, India
Keywords: AdaBoost Regressor, Cars, Forecast, Machine Learning, Novel XGBoost Regressor, Vehicles.
Abstract: This study primarily aimed to estimate used car prices using the XGBoost Regressor to enhance accuracy in
relation to vehicle data and benchmark this against the AdaBoost Regressor. The proposed methodology
assessed the XGBoost and AdaBoost Regressor algorithms, represented by 10 samples split into two sets.
This assessment employed an 80% Gpower and a 95% confidence interval. It emerged that the Novel
XGBoost Regressor achieved an accuracy of 87.74%, surpassing the AdaBoost Regressor's 84.31%.
Furthermore, SPSS statistical analysis confirmed the significance of both algorithms with p=0.000 (p<0.05).
Ultimately, the results highlight the superior accuracy of the Novel XGBoost Regressor over the AdaBoost
Regressor in predicting used car prices.
1 INTRODUCTION
The primary aim of this research is to develop
machine learning models that accurately predict used
car prices based on specific parameters or features.
The dataset utilised comprises the selling prices of
various car models from different locations (Yadav,
Kumar, and Yadav 2021; Ramkumar G et al. 2021).
As the number of private vehicles rises and the used
car market evolves, used cars are becoming
increasingly vital for buyers. For both parties,
understanding the price of a used car is essential for a
smooth transaction. Knowing the cost of used
vehicles empowers buyers to negotiate confidently,
while determining the residual value aids sellers in
setting reasonable prices (Liu et al. 2022). The
necessity arises as personal vehicles are integral for
commuting between homes and workplaces. The
decision to purchase a new car can be daunting, but
deciding how to price a current car for sale can be
even more challenging. The high cost of new vehicles
often restricts consumer buying power. Various
methods exist to estimate a car's price based on its
market value (Asghar et al. 2021; Palanivelu et al.
2022). Machine learning could offer a solution. Here,
regression algorithms predict used car selling prices
using the Python module, Scikit-Learn, relying on
*
Research Scholar
Research Guide, Corresponding Author
historical car data (Vickram et al. 2016; C. Chen,
Lulu Hao, and Congfu Xu 2017). Regarding used car
price prediction, numerous studies have been
published over the past five years. Springer Link
released about 3822 articles, Google Scholar had
1690 related papers, ResearchGate published 40
pieces, and IEEE Xplore featured 31 articles. A
notable study by Narayana et al. (2021) introduced a
fair pricing system, forecasting vehicle prices based
on features like model, age, fuel type, and seller type.
AoQiang Wang et al. (2022) categorised the used car
transaction cycle using six machine learning models
and subsequently identified crucial factors
influencing used car transactions. Gültek
n and Organ
(2020) employed a method known as ANN to predict
vehicle prices in the used car market. Other studies
have compared the efficacy of regression using
supervised machine learning models and developed
mathematical models for price prediction based on
car attributes (Nitis Monburinon et al. 2018; Santosh
Kumar Satapathy, Rutvikraj Vala, and Shiv Virpariya
2022). Estimating a used car's value is complex due
to various influencing factors, such as its condition,
mileage, year, and make.
A research gap has been identified, highlighting a
lack of accuracy in current research regarding used
car price predictions. This study aims to enhance
Reddy, S. and Kumar, S.
A Comparative Analysis of XGBoost Model and AdaBoost Regressor for Prediction of Used Car Price.
DOI: 10.5220/0012510700003739
Paper published under CC license (CC BY-NC-ND 4.0)
In Proceedings of the 1st International Conference on Artificial Intelligence for Internet of Things: Accelerating Innovation in Industry and Consumer Electronics (AI4IoT 2023), pages 441-446
ISBN: 978-989-758-661-3
Proceedings Copyright © 2024 by SCITEPRESS – Science and Technology Publications, Lda.
441
accuracy by deploying the Novel XGBoost
Regressor, contrasting its performance with the
AdaBoost Regressor.
2 MATERIALS AND METHODS
This research was conducted in the Lab of Artificial
Intelligence at the Saveetha School of Engineering,
part of the Saveetha Institute of Medical and
Technical Sciences in Chennai. The study comprised
two groups, each with a sample size of 10 sets,
making a total of 20 sample sets considered for this
research (Hankar Mustapha et al. 2022). Sample
estimation was carried out with an alpha value of
0.05, a statistical power of 80%, and a 95%
confidence interval. The sample size was determined
using the ClinCalc software under supervised
learning conditions. The tools utilised for this study's
statistical analysis were version 26.0.1 of the SPSS
software and the Jupyter Notebook.
Vehicle data was sourced from the well-known
platform Kaggle.com, specifically from a dataset made
available in Vijayaadithyan's 2022 Kaggle repository.
This dataset includes 18 attributes, such as id, region,
manufacturer, model, condition, odometer, and so
forth. After the initial data collection, all missing and
null values found within the vehicle dataset were
eliminated through data cleaning and preprocessing
steps. Ultimately, the algorithms were compared to
determine which one was more effective.
2.1 Novel XGBoost Regressor
Extreme Gradient Boosting (XGBoost) is a renowned
machine learning algorithm, an enhanced and
optimised version of the gradient boosting method,
crafted for efficiency, speed, and model quality.
Owned by the Distributed Machine Learning
Community, this open-source library aims to refine
and expedite established boosting techniques. It
harmoniously integrates hardware and software
capabilities. In our study, we use classification
boosting with XGBoost. The boosting algorithm
successively generates weak learner models and
amalgamates their predictions to augment overall
model performance. When an erroneous prediction
occurs, misclassified samples are attributed higher
weights, while correctly classified ones receive lower
weights. The resulting ensemble model assigns
greater weights to better-performing weak learner
instead, it adjusts subsequent ones by learning from
past errors. Given boosting's greedy nature, it's
prudent to set stopping criteria, such as model
performance (early stopping) or specific iterations,
like tree depth for tree-based learners, to evade
overfitting the training data.
2.2 Procedure for the Novel XGBoost
Regressor
1. Initialise required modules and libraries.
2. Import the target dataset.
3. Undertake preprocessing tasks, encompassing
data cleaning and transformation.
4. Execute feature engineering to refine the
dataset and follow up with exploratory data
analysis (EDA) to gain insights.
5. Partition the dataset into training and testing
subsets.
6. Validate the divided data to ensure its quality
and distribution.
7. Deploy the Novel XGBoost Regressor to train
on the training set and subsequently test it.
8. Evaluate and display the model's accuracy.
2.3 AdaBoost Regressor
Adaptive Boosting, commonly known as AdaBoost,
is a renowned ensemble learning method in the realm
of machine learning. Primarily effective for binary
classification tasks, AdaBoost can also be adapted for
multi-class classification or regression challenges.
Typically, AdaBoost employs a one-level decision
tree, often termed a "decision stump". The name
"stump" stems from its simplistic nature, signifying a
tree cut down to just a trunk.
The foundational concept of AdaBoost is
"boosting". Its mechanism is delineated as follows:
a. Develop the inaugural predictive model using
the original dataset.
b. Construct a subsequent predictive model that
rectifies the shortcomings of the initial model.
c. Generate a third model addressing the
imperfections of the second.
d. Continue creating models until either the set
maximum number of iterations (or trees) is
met or the model achieves absolute accuracy.
Procedure For Adaboost Regressor
1. Load the essential libraries.
2. Load the dataset.
3. Identify the X (features) and Y (target)
variables.
4. Conduct preprocessing on the dataset,
encompassing data cleaning and data
transformation.
AI4IoT 2023 - First International Conference on Artificial Intelligence for Internet of things (AI4IOT): Accelerating Innovation in Industry
and Consumer Electronics
442
5. Split the dataset into training and testing sets.
6. Implement the AdaBoost Regressor model
and fit the data.
7. Evaluate the model's performance.
8. Display the accuracy.
3 STATISTICAL ANALYSIS
The SPSS software was used for the descriptive
statistical analysis of both the Novel XGBoost and
AdaBoost Regressor models. In this dataset, the year
and odometer are considered as independent
variables, while lat, long, and price are treated as
dependent variables. An independent samples t-test
was employed, as cited by Cui et al. (2022).
4 RESULTS
The performance of the Novel XGBoost Regressor
appears to be notably superior to that of the AdaBoost
Regressor. As depicted in Table 1, the accuracy of the
Novel XGBoost Regressor stands at 87.74%, whereas
the AdaBoost Regressor achieved an accuracy of
84.31%.
Table 2 provides the group statistics of both the
Novel XGBoost Regressor and the AdaBoost
Regressor, highlighting a statistically significant
difference between the two.
Table 1: Accuracy of Novel XGBoost Regressor and
AdaBoost Regressor.
S.NO
Novel XGBoost
Regressor
AdaBoost Regressor
1
89.45
86.40
2
86.37
85.32
3
87.50
83.66
4
88.10
84.20
5
89.00
81.58
6
87.00
83.36
7
85.53
84.53
8
86.70
82.15
9
89.20
85.71
10
88.60
86.22
The accuracy is utilised to gauge the overall
performance in predicting used car prices.
In Figure 1, the Standard Deviation and Mean
Accuracy for both algorithms are displayed: for the
Novel XGBoost Regressor, they are 1.33167 and
87.74% respectively, and for the AdaBoost
Regressor, 1.64992 and 84.31%. The figure contrasts
the two algorithms (Novel XGBoost Regressor vs
AdaBoost Regressor) on the X-Axis, while the Y-
Axis showcases the Detection of Mean Accuracy,
considering a range of +/- 2SD.
Table 2: Group Statistics of XGBoost Regressor (Mean Accuracy of 87.74%) AdaBoost Forest Regressor (Mean Accuracy
of 84.31%).
Algorithm
N
Mean Accuracy
Std.Deviation
Std. Error Mean
Novel XGBoost
Regressor
10
87.74
1.33167
.42111
AdaBoost Regressor
10
84.31
1.64992
.52175
Table 3: Independent Sample Test T-test is applied for the data set fixing the confidence interval as 95%. It shows that there
is a statistically significant difference between XGBoost Regressor and AdaBoost Regressor with a two-tailed value p=0.000
(p<0.05).
for Equality of
Variances
t-test for Equality of Means
F
Sig
t
df
Sig(2-
tailed)
Mean
Difference
Std. Error
Difference
95% Confidence Interval
of the Difference
Lower
Upper
Accuracy
Equal
variances
assumed
.343
.566
5.119
18
.000
3.432
.67049
2.02335
4.84065
Equal
variances not
assumed
5.119
17.232
.000
3.432
.67049
2.01884
4.84516
A Comparative Analysis of XGBoost Model and AdaBoost Regressor for Prediction of Used Car Price
443
Figure 1: Comparison of Novel XGBoost Regressor and AdaBoost Regressor using a graphical representation. And the Mean
Accuracy and Standard Deviation for Novel XGBoost Regressor (87.74% and 1.33167) and AdaBoost Regressor (84.31%
and 1.64992). X-Axis: Novel XGBoost Regressor vs AdaBoost Regressor algorithms and Y-Axis: Detection of Mean
accuracy +/- 2SD.
Lastly, Table 3 illustrates the categories
associated with the Independent samples test, the test
for equality of means, standard error differences, and
the test for equality of variances. This table
encompasses the Mean of Standard Error, Group
statistics for Mean and Standard Deviation, each with
a sample size of 10. The findings underscore that both
the proposed method (XGBoost Regressor) and the
comparative method (AdaBoost Regressor) have
statistical significance, with a p-value of 0.000
(p<0.05), based on the statistical analysis conducted
via SPSS.
5 DISCUSSIONS
The aforementioned study showcases the superior
predictive accuracy of the Novel XGBoost Regressor
in determining used car prices, standing at 87.74%,
when compared to the AdaBoost Regressor, which
demonstrates an accuracy of 84.31%. Enhanced
performance was observed with the algorithm when
exposed to a larger volume of training data. An
Independent samples T-test analysis highlighted a p-
value of 0.000 (p<0.05), signifying the statistical
significance of both the proposed and comparative
algorithms.
In alignment with these findings, several research
publications have reinforced the outcomes of this
study. One research piece by K. Samruddhi and R. A.
Kumar in 2020 employed various training and testing
ratios to examine the data, ultimately recommending
an optimal model boasting an accuracy close to 85%.
Pal et al. (2019) proposed the use of the Random
Forest algorithm, attaining an accuracy of 83.63%.
Experimental results from a study by Janke, K, and C
in 2022 illustrated the superior performance of the
Random Forest model relative to other models,
evidenced by its R^2 score of 0.772584 (translating
to an accuracy of 77.2%) and a Mean Absolute Error
value of 1.0970472. Bharambe et al. (2022)
implemented lasso regression, introducing a model
with a commendable accuracy of 87%. Notably, no
research articles have been found that contest the
proposed algorithm's efficacy for predicting used car
prices in the context of this study.
However, the study is not without its limitations.
Recent pandemic-induced semiconductor shortages,
which inadvertently augmented the price of used cars,
are pinpointed as one of the research's primary
constraints. Owing to the sudden shift in car pricing
amidst the study's tenure, the current dataset may not
aptly represent market-available vehicles, making it
arduous to predict future car costs with precision. In
the trajectory of future research, the incorporation of
advanced machine learning techniques can be
contemplated to elevate the model's accuracy and
optimization levels. A real-time data model,
seamlessly integrable within a mobile application and
boasting widespread usage, might emerge as an
optimal solution.
AI4IoT 2023 - First International Conference on Artificial Intelligence for Internet of things (AI4IOT): Accelerating Innovation in Industry
and Consumer Electronics
444
6 CONCLUSION
In the ongoing quest for developing optimal machine
learning algorithms for predicting used car prices, the
Novel XGBoost Regressor has emerged as a notably
effective solution. When compared against the
AdaBoost Regressor, the Novel XGBoost Regressor
outperforms with a commendable accuracy rate of
87.74%. In contrast, the AdaBoost Regressor lags
slightly behind with an accuracy of 84.31%.
Several aspects from the study underline this
superiority:
• Statistical Significance: A rigorous T-test
analysis for independent samples was
conducted, leading to a p-value of 0.000
(p<0.05). This result alone provides a
compelling argument for the statistical
superiority of the Novel XGBoost Regressor
over the AdaBoost Regressor. The low p-
value highlights that the difference in their
performances is significant and not just a
random occurrence.
• Supporting Literature: Multiple research
endeavors in the domain corroborate the
findings. For instance, a study by K.
Samruddhi and R. A. Kumar in 2020 reached
a model accuracy close to 85%. Another
investigation using the Random Forest
algorithm, as suggested by Pal et al. (2019),
secured an accuracy of 83.63%. These
findings align with the current study's
observations that advanced machine learning
techniques, like the Novel XGBoost
Regressor, tend to outdo traditional methods
in predicting used car prices.
• Mean Accuracy and Standard Deviation: The
analysis further delineated the Standard
Deviation and Mean Accuracy of both models.
While the AdaBoost Regressor recorded a
mean accuracy of 84.31% with a standard
deviation of 1.64992, the Novel XGBoost
Regressor surpassed it with a mean accuracy
of 87.74% and a slightly lower standard
deviation of 1.33167. This not only highlights
the higher accuracy of the Novel XGBoost but
also suggests its consistent performance
across different datasets.
• Limitations & Future Scope: Despite its
impressive accuracy, the study did confront
challenges linked to recent economic
phenomena, such as the semiconductor
shortage triggered by the pandemic, which
affected used car prices. The dataset's
potential inability to represent real-time
market prices underscores the necessity for
continuous algorithm updating and training.
In conclusion, while both models hold merit in
their respective capacities, the Novel XGBoost
Regressor exhibits a clear edge in the context of
predicting used car prices. Its amalgamation of speed,
efficiency, and accuracy makes it a potent tool for
researchers and industry professionals, offering a
blueprint for future advancements in the realm of
machine learning-powered price predictions.
REFERENCES
Santosh Kumar Satapathy, Rutvikraj Vala, and Shiv
System Using Effective Machine Learning
https://doi.org/10.1109/CISES54857.2022.9844350.
Asghar, Muhammad, Khalid Mehmood, Samina Yasin, and
Prediction Using Machine Learning with Optimal
Pakistan Journal of Engineering and
Technology 4 (2): 11319.
Bharambe, Prof Pallavi, Bhargav Bagul, Shreyas Dandekar,
International Journal for Research in Applied Science
and Engineering Technology 10 (4): 77378.
Cui, Baoyang, Zhonglin Ye, Haixing Zhao, Zhuome
Price Prediction Based on the Iterative Framework of
Electronics 11 (18): 2932.
Gültek
Secondhand Cars Sold on the Internet with Artificial
Journal of Internet
Applications and Management 11 (1): 4961.
Janke, Varshitha, Jahnavi K, and Lakshmi C. 2022.
2022
International Conference on Computer Communication
and Informatics (ICCCI). IEEE.
https://doi.org/10.1109/iccci54379.2022.9740817.
Liu, Enci, Jie Li, Anni Zheng, Haoran Liu, and Tao Jiang.
Car Price in View of the PSO-GRA-BP Neural
Sustainability: Science Practice and Policy
14 (15): 8993.
Kishore Kumar, M. Aeri, A. Grover, J. Agarwal, P. Kumar,
10.1109/icesc51422.2021.9532845.
Palanivelu, J., Thanigaivel, S., Vickram, S., Dey, N.,
Mihaylova, D., & Desseva, I. (2022). Probiotics in
functional foods: survival assessment and approaches
for improved viability. Applied Sciences, 12(1), 455.
Pal, Nabarun, Priya Arora, Puneet Kohli, Dhanasekar
Sundararaman, and Sai Sumanth Palakurthy. 2019.
A Comparative Analysis of XGBoost Model and AdaBoost Regressor for Prediction of Used Car Price
445
Advances in Information and Communication
Networks, 41322.
Nitis Monburinon, Prajak Chertchom, T. Kaewkiriya,
Suwat Rungpheung, Sabir Buya, and Pitchayakit
https://doi.org/10.1109/ICBIR.2018.8391177.
AoQiang Wang, Qiancheng Yu, Xiaoning Li, Zekun Lu,
Used Car Valuation Problem Based on Machine
https://doi.org/10.1109/ICCNEA57056.2022.00032.
for Analyzing the Mechanical Properties of Natural
Fiber Composite Article ID 5450935, 15 pages, 2021.
https://doi.org/10.1155/2021/5450935
https://www.kaggle.com/vijayaadithyanvg/car-
price-predictionused-cars.
V. P. Parandhaman, "An Automated Efficient and Robust
Scheme in Payment Protocol Using the Internet of
Things," 2023s, pp. 1-5, doi:
10.1109/ICONSTEM56934.2023.10142797.
Mustapha Hankar, Marouane Birjali, and Abderrahim Beni-
.
http://dx.doi.org/10.1109/ISIVC54825.2022.9800719.
Vickram, A. S., Kamini, A. R., Das, R., Pathy, M. R.,
Parameswari, R., Archana, K., & Sridharan, T. B.
(2016). Validation of artificial neural network models
for predicting biochemical markers associated with
male infertility.
Chuancan Chen, Lulu Hao, and Cong Xu. 2017.
www.doi.org.10.1063/1.4982530.
Price Prediction using K-Nearest Neighbor Based
www.doi.org.10.29027/IJIRASE.v4.i2.2020.629-632.
Yadav, Anu, Ela Kumar, and Piyush Kumar Yadav. 2021.
Analysis System (UCPAS) Using Machine Learning
Linguistics and Culture Review
AI4IoT 2023 - First International Conference on Artificial Intelligence for Internet of things (AI4IOT): Accelerating Innovation in Industry
and Consumer Electronics
446
