Comparison of Enhanced XGBoost Algorithm with Light Gradient

Boosting Machine to Determine the Prediction of Black Friday Sales

Koyyala Ramprasad

and R. Thalapathi Rajasekaran

†

Department of Computer Science and Engineering, Saveetha School of Engineering, Saveetha Institute of Medical and

Technical Sciences, Saveetha University, Chennai, Tamil Nadu, India

Keywords: Black Friday Sales, Forecast, Light Gradient Boosting Machine, Machine Learning, Enhanced XGBoost,

Retail Industry.

Abstract: This research evaluates the efficacy of the Enhanced XGBoost algorithm in forecasting sales during the Black

Friday event, juxtaposed against the Light Gradient Boosting Machine's performance. With a focus on

enhancing sales prediction precision, the study uses both the Enhanced XGBoost and the Light Gradient

Boosting Machine. Employing a sample size of 10 for each, determined using ClinCalc software with a

confidence interval of 95% and an alpha value of 0.05, the investigation relies on a sales analysis dataset from

Kaggle, comprising 80,550 entries. SPSS statistical analysis reveals the Enhanced XGBoost's accuracy stands

at 93%, outstripping the Light Gradient Boosting Machine's 73%. Furthermore, a significant difference is

observed between the two, with a p-value of 0.001. The findings clearly show the Enhanced XGBoost's

superior performance in Black Friday sales predictions.

1 INTRODUCTION

The significance of this subject lies in predicting

Black Friday sales, thus aiding retail businesses in

crafting bespoke offers and promotions for their

customers. In the UK, Black Friday denotes the day

after Thanksgiving, marking one of the most hectic

shopping days in the retail calendar. Americans

celebrate Thanksgiving annually on the third

Thursday of November (Habel, Alavi, and Heinitz

2022). Though the exact origins of 'Black Friday'

remain uncertain, popular legend suggests that traffic

officers in central Philadelphia first used the term in

1965 to describe the throngs of shoppers and resultant

traffic (Lin 2020). By the 1980s, retailers had

embraced the term, using it to allude to accounting

practices where profits were marked in black ink and

losses in red (Sohan and SAARIKA 2020; Vickram

et al. 2021). The goal of predicting Black Friday sales

is to forecast future sales and fathom customer

behaviour, utilising an extreme gradient boosting

regression algorithm for enhanced price accuracy

compared to the Light Gradient Boosting Machine

(Wang 2021). Upon reflection, scholars have

predicted Black Friday sales, contributing 80 relevant

Research Scholar

†

Research Guide, Corresponding Author

research articles to IEEE Digital Xplore, 40 articles in

Science Direct, 90 in Google Scholar, and a

remarkable 5,305 in SpringerLink (Arora 2021; AS et

al. 2013). The term "Black Friday" implies the day

retailers endeavour to turn their losses (denoted as

"red") into profits ("black"). Regardless of its origins,

the monumental influence of this singular shopping

day on the retail sector is undeniable. In 2008, Black

Friday sales soared to unprecedented heights, seeing

consumers splurge $10.6 billion in just 24 hours

(Saura, Reyes-Menendez, and Palos-Sanchez 2019).

In a bid to entice shoppers, stores in recent years have

begun opening their doors at midnight on Black

Friday (Petroșanu 2022). Due to this shift, strategies

that reward the earliest shoppers have emerged. The

optimal review on Black Friday sales forecasting,

employing machine learning techniques, is housed in

Google Scholar (Manko and Jose 2022; Ramkumar

G. et al. 2021).

Existing research confronts issues, chiefly the

imprecision in forecasting Black Friday sales. In this

context, the Enhanced XGBoost Algorithm and Light

Gradient Boosting Machine are employed to predict

Black Friday sales in retail (Nguyen 2019). This

assists consumers in making well-informed

474

Ramprasad, K. and Rajasekaran, R.

Comparison of Enhanced XGBoost Algorithm with Light Gradient Boosting Machine to Determine the Prediction of Black Friday Sales.

DOI: 10.5220/0012518100003739

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 474-479

ISBN: 978-989-758-661-3

purchasing decisions and business owners in setting

product prices that reflect its value. This concept of

sales forecasting equips individuals to make informed

choices when shopping in the future. Therefore, the

crux of this work is predicting sales accuracy during

Black Friday using the Enhanced XGBoost

Algorithm, juxtaposed against the Light Gradient

Boosting Machine.

2 MATERIALS AND METHODS

The research was conducted in the Artificial Studio at

Saveetha School of Engineering. There are two

primary groups involved. In the first group, the

Enhanced XGBoost Regression is employed, while

the second group uses the Light Gradient Boosting

Machine Regression. The sample size for each group

is set at 10, as determined by the ClinCalc software

(Hoel 2022). The calculations involve an 80% G-

power, with an alpha value set at 0.05 and a 95%

confidence interval. The study assesses the accuracy

of two algorithms, the Enhanced XGBoost

Regression and LightGBM. The input dataset for this

research is the Sales analysis dataset, sourced from

the Kaggle repository (Lopes 2022), containing a

total of 80,550 entries. The dataset features

independent variables like occupation and

city_category, with the purchase acting as the

dependent variable. The dataset comprises twelve

attributes: Gender, Age, Occupation, City_Category,

Stay_In_Current_City_Years, Martial_Status,

Product_Category_1 through Product_Category_4,

and Purchase. For the study, 10 samples are drawn for

each group. The dataset is then bifurcated into

training and testing categories; the training data aids

in forecasting accuracy.

2.1 Enhanced XGBoost Regression

Enhanced XGBoost, or Extreme Gradient Boosting,

is rooted in the ensemble technique known as

'boosting'. This technique involves iteratively adding

models to address the shortcomings or errors in

preceding ones. The process continues until no

substantial improvements are observed. Gradient

boosting, a subset of this technique, involves creating

models to predict the errors or residuals of prior

models. These models are then amalgamated for the

final prediction. The 'gradient' in gradient boosting

arises from its use of gradient descent methodologies

to minimise the loss as new models are incorporated.

The Enhanced XGBoost library prioritises both

model efficacy and computational efficiency. It's

crafted to optimise memory use and ensure quick

computational processes.

Here's the revised version of the procedure list using

British English:

Procedure

1. Understand the problem statement.

2. Familiarise oneself with the dataset.

3. Analyse the features and describe the

categorical ones.

4. Describe the numerical features. There are two

types of numerical features: discrete and

continuous.

5. Describe the relationship between the

independent and dependent variables.

6. Check for correlations among independent

variables.

7. Convert categorical features to numerical ones.

8. Select features with precision.

9. Divide the dataset into training, testing, and

validation subsets.

10. Set the hyperparameters for Enhanced

XGBoost regression.

11. Fit the data using x-train, y-train, x-test, and y-

test.

12. Predict accuracy.

13. Compare validation accuracy with overall

accuracy.

14. If accuracy is unsatisfactory, retrain the data

using different independent features.

15. Choose the best accuracy.

2.2 Light Gradient Boosting Machine

Microsoft developed the distributed gradient boosting

algorithm for a machine learning framework called

LightGBM, also known as the Light Gradient

Boosting Machine. The "Light" in LightGBM

denotes its capacity for swift calculations. It can

manage vast quantities of data and operates with

minimal memory consumption. It's especially

suitable for large datasets where high precision is

essential. LightGBM is a gradient-boosting

framework that utilises the decision tree model in

machine learning to enhance model performance,

reduce training time, and minimise memory usage.

LightGBM can achieve faster speeds without

compromising on accuracy. It supports both

classification and regression tasks. This gradient-

boosting ensemble technique, based on decision trees,

is facilitated by the AutoML tool.

Procedure

1. Understand the problem statement and

familiarise oneself with the dataset.

Comparison of Enhanced XGBoost Algorithm with Light Gradient Boosting Machine to Determine the Prediction of Black Friday Sales

475

2. Examine the attributes and characterise the

categorical features.

3. Define the relationship between the

independent and dependent variables.

4. Assess the correlation of the independent

variable.

5. Convert categorical features to numerical

ones.

6. Select the most suitable features.

7. Split the dataset into training, testing, and

validation sets.

8. Set hyperparameters for LGBM regression.

9. Fit the models using x-train, y-train, x-test,

and y-test.

10. Predict and evaluate accuracy.

11. Compare validation accuracy with overall

accuracy.

12. If accuracy is not satisfactory, retrain the data

using different independent features.

13. Determine the optimal approach.

3 STATISTICAL ANALYSIS

In the analysis, descriptive statistics for the Enhanced

XGBoost Regressor and Light Gradient Boosting

Machine Regressor were conducted using SPSS. In

this dataset, the independent variables include

Occupation and City_Category, while the dependent

variable is Purchase. Tian (2022) employed an

independent-samples t-test.

4 RESULTS

Table 1 displays the accuracy values from iterations

of the Enhanced XGBoost Regressor at 93.000% and

the Light Gradient Boosting Machine Regressor at

73.000%. The Enhanced XGBoost Regressor appears

to outperform the Light Gradient Boosting Machine

Regression slightly.

Table 2 provides the group statistics. The

Enhanced XGBoost Regressor achieves an average

accuracy of 93.00% with a standard deviation of

0.03722. In contrast, the Light Gradient Boosting

Machine Regressor has a standard deviation of

0.4219 and an average accuracy of 73.00%. The

results indicate that the Enhanced XGBoost

Regressor performance surpasses that of the Light

Gradient Boosting Machine Regressor.

Table 3 presents the independent sample T-test

values for the Enhanced XGBoost Regressor and the

Light Gradient Boosting Machine Regressor,

showing a mean difference of 18.53747 and a

standard error difference of 2.06520. A statistically

significant difference exists between the Enhanced

XGBoost algorithm and the Light Gradient Boosting

Machine algorithm, as evidenced by a 2-tailed p-

value of 0.001 (p<0.05).

Table 1: Accuracy Values for XGB (93.00) and LGBM (73.00).

NUMBER OF ITERATIONS

XGB

LGB

93.00

73.00

93.05

73.10

92.99

72.95

92.95

72.99

93.12

73.00

93.00

73.08

93.00

72.95

93.00

73.00

92.99

72.99

92.92

72.94

Table 2: The mean and standard deviation of the group and accuracy of the Enhanced XGBoost and Light Gradient Boosting

Machine algorithms where XGB has a mean accuracy (93.00), std. deviation (0.03722), whereas Light Gradient Boosting

Machine Regressor has a mean accuracy (73.00), std. deviation (0.4219).

Group Statistics

Accuracy

Group

Mean

Std.Deviation

Std. Error Mean

XGB

93.0000

0.3722

0.1241

LGBM

73.0000

0.4219

0.1673

AI4IoT 2023 - First International Conference on Artiﬁcial Intelligence for Internet of things (AI4IOT): Accelerating Innovation in Industry

and Consumer Electronics

476

Table 3: An Independent sample T-test can evaluate the significance and standard error for two groups. There is a statistical

significance difference between the Enhanced XGBoost algorithm and Light Gradient Boosting Machine algorithm with p

value of p=0.001 (p<0.05).

Independent Samples Test

Levene’s Test for Equality of Variances

T-test for Equality of Means

Sig

Sig(2-

tailled)

Mean

Difference

Std.error

diff

95% Confidence

Interval of the diff

Lower

Upper

Accuracy

Equal

variances

assumed

3.90

0.064

8.976

0.001

18.53747

2.065

14.19

22.87

Equal

variances

not

assumed

8.973

10.0

0.001

18.53747

1.858

14.39

22.67

Figure 1: Comparison of Accuracy of Enhanced XGBoost Regression with Light Gradient boosting Regression in terms of

mean Accuracy. The mean accuracy of the Enhanced XGBoost Regression is higher than the Light Gradient boosting

Regression. Bar graph comparison of Enhanced XGBoost regression with Light gradient boosting Regression in Sales

Prediction. X-axis algorithm, Y-axis mean accuracy with +/- 2 SD.

Figure 1 depicts a bar graph comparing the

average accuracy of both the Enhanced XGBoost

Regressor and the Light Gradient Boosting Machine

Regressor algorithms. The Enhanced XGBoost

Regressor has a mean detection accuracy of 93.00%,

while the Light Gradient Boosting Machine

Regressor stands at 73.000%, both measured with a

variance of +/- 2S.

5 DISCUSSION

In this study, the prediction of Black Friday sales

using the Enhanced XGBoost regression method

yielded a considerably higher accuracy of 93%

compared to the Light Gradient Boosting Machine

Regression, which achieved an accuracy of 73%. The

results from the Enhanced XGBoost appear more

consistent and showcase a lower standard deviation.

The current methodology for Black Friday sales

prediction employs the Light Gradient Boosting

Machine Regressor, whereas the proposed approach

integrates the Enhanced XGBoost algorithms in

Machine Learning (Tiainen 2021). The Enhanced

XGBoost boasts an accuracy of 93%, making it adept

at predicting Black Friday sales based on various

features and specifications (Liu 2022). Achieving the

desired accuracy often stems from enhancing the

quality of the dataset. The Enhanced XGBoost's

performance excels further when integrated with

other machine-learning algorithms. The primary aim

of these promotional efforts is to entice consumers to

increase their online purchases, thereby bolstering the

e-commerce and retail sectors (Jandera and

Skovranek 2022). To both predict and train, a suitable

dataset is essential. The most expansive online dataset

for this purpose is named the Black Friday Sales

Dataset (Moon, Park, and Kim 2019).

Comparison of Enhanced XGBoost Algorithm with Light Gradient Boosting Machine to Determine the Prediction of Black Friday Sales

477

This study's limitations include potential issues

arising when the number of predictions falls short of

the number of observations. If there's a presence of

two or more highly collinear variables, one is chosen

arbitrarily. The method is susceptible to over-

amplification, and insufficient data might not deliver

the anticipated results. Furthermore, not all the

relevant training factors are considered. Looking

forward, the planned work's subsequent phase will

centre on forecasting Black Friday sales using

regression techniques and further refining the

accuracy of the Enhanced XGBoost.

6 CONCLUSION

In the constantly evolving realm of predictive

analytics, especially in the retail sector, accurate

forecasting remains a cornerstone for effective

decision-making. Black Friday, a crucial sales event

for retailers, necessitates precision in predictions to

ensure they remain competitive and meet consumer

demand. In light of our recent research, we assessed

two prominent predictive algorithms: the Enhanced

XGBoost Regression and the Light Gradient

Boosting Machine Regressor, to determine their

efficacy in forecasting Black Friday sales.

Here are six key points stemming from our

analysis:

● Performance Consistency: The Enhanced

XGBoost Regression not only achieved a higher

accuracy but also showed consistent results across

various test scenarios.

● Standard Deviation: The consistency of the

Enhanced XGBoost Regression is further

exemplified by its smaller standard deviation,

indicating lesser variability in its predictions.

● Adaptability to Data: The Enhanced XGBoost

Regression exhibited superior adaptability to

different datasets, making it a more versatile tool

for diverse prediction scenarios.

● Computational Efficiency: Although not

explicitly mentioned earlier, it's worth noting that

algorithms like XGBoost are often designed for

optimized computational efficiency, which can be

crucial for large datasets.

● Collinearity Handling: One inherent strength of

Enhanced XGBoost Regression is its ability to

manage collinear variables more effectively

compared to some other algorithms.

● Integration with Other Tools: The modularity of

the Enhanced XGBoost Regression allows for its

seamless integration with other machine learning

algorithms, enhancing its predictive capabilities

further.

In conclusion, our analysis clearly indicates the

superior performance of the Enhanced XGBoost

Regression over the Light Gradient Boosting

Machine Regressor in predicting Black Friday sales.

While the latter algorithm has an accuracy of 73%,

the former impressively scores 93%. This disparity in

accuracy underscores the potential of the Enhanced

XGBoost Regression, rendering it a more reliable tool

for precise forecasting in the retail industry,

especially for monumental sales events like Black

Friday.

REFERENCES

Arora, Ashish, Bhupesh Bhatt, Divyanshu Bist, Rachna

Jain, and Preeti Nagrath (2021). “Predicting Customer

Spent on Black Friday.” Proceedings of Second

International Conference on Computing,

Communications, and Cyber-Security, 183–201.

AS, Vickram, Raja Das, Srinivas MS, Kamini A. Rao, and

Sridharan TB (2013). "Prediction of Zn concentration

in human seminal plasma of Normospermia samples by

Artificial Neural Networks (ANN)." Journal of assisted

reproduction and genetics 30: 453-459.

Habel, Johannes, Sascha Alavi, and Nicolas Heinitz (2022).

“A Theory of Predictive Sales Analytics Adoption,”

December. https://doi.org/10.2139/ssrn.3994561.

Jandera, Ales, and Tomas Skovranek (2022). “Customer

Behavior Hidden Markov Model.” Science in China,

Series A: Mathematics 10 (8): 1230.

Kishore Kumar, M. Aeri, A. Grover, J. Agarwal, P. Kumar,

and T. Raghu, (2023) “Secured supply chain

management system for fisheries through IoT,” Meas.

Sensors, doi: 10.1016/j.measen.2022.100632.

Lin, Chuyang, Yiwei Huang, Yibing Tan, Xiaocun Lu,

Yujie Zhou, Yifei Liu, Paul Wescott, and Sasha Stoikov

(2020).” February.

https://doi.org/10.2139/ssrn.3533358.

Liu, Zekun, Weiqing Zhang, Xiao Liu, Eitan Muller, and

Feiyu Xiong (2022). “Success and Survival in

Livestream Shopping,” February.

https://doi.org/10.2139/ssrn.4028092.

Lopes, Gustavo (2022). “The Wisdom of Crowds in

Forecasting at High-Frequency for Multiple Time

Horizons: A Case Study of Brazilian Retail Sales.”

Brazilian Review of Finance 20 (2): 77–115.

Manko, Barbara A., and Susheel Tom Jose (2022).

“Gender-Based Shopping And Consumer Purchasing:

‘Sale Ends Today.’” European Journal of Management

and Marketing Studies 7 (4).

https://doi.org/10.46827/ejmms.v7i4.1299.

Moon, Sangkil, Yoonseo Park, and Yong Seog Kim (2019).

“The Impact of Text Product Reviews on Sales.”

European Journal of Marketing 48 (11/12): 2176–97.

AI4IoT 2023 - First International Conference on Artiﬁcial Intelligence for Internet of things (AI4IOT): Accelerating Innovation in Industry

and Consumer Electronics

478

Petroșanu, Dana-Mihaela, Alexandru Pîrjan, George

Căruţaşu, Alexandru Tăbușcă, Daniela-Lenuța Zirra,

and Alexandra Perju-Mitran (2022). “E-Commerce

Sales Revenues Forecasting by Means of Dynamically

Designing, Developing and Validating a Directed

Acyclic Graph (DAG) Network for Deep Learning.”

Electronics 11 (18): 2940.

Ramkumar, G. et al (2021). “An Unconventional Approach

for Analyzing the Mechanical Properties of Natural

Fiber Composite Using Convolutional Neural

Network” Advances in Materials Science and

Engineering vol. 2021, Article ID 5450935, 15 pages,

2021. https://doi.org/10.1155/2021/5450935

Saura, Jose Ramon, Ana Reyes-Menendez, and Pedro

Palos-Sanchez (2019). “Are Black Friday Deals Worth

It? Mining Twitter Users’ Sentiment and Behavior

Response.” Journal of Open Innovation: Technology,

Market, and Complexity 5 (3): 58.

Sohan, A., and SAARIKA (2020). “An Analysis of

Consumer Behaviour on Black Friday,” September.

http://hdl.handle.net/123456789/12681.

Tiainen, Matias (2021). “Forecasting Seasonal Demand at

the Product Level in Grocery Retail.”

https://aaltodoc.aalto.fi/handle/123456789/107631.

Tian, Liwei, Li Feng, Lei Yang, and Yuankai Guo (2022).

“Stock Price Prediction Based on LSTM and

LightGBM Hybrid Model.” The Journal of

Supercomputing 78 (9): 11768–93.

Vickram, A. S., Srikumar, P. S., Srinivasan, S., Jeyanthi, P.,

Anbarasu, K., Thanigaivel, S., ... & Rohini, K (2021).

Saudi journal of biological sciences, 28(6), 3607-3615.

V. P. Parandhaman, (2023). “An Automated Efficient and

Robust Scheme in Payment Protocol Using the Internet

of Things," doi:

10.1109/ICONSTEM56934.2023.10142797.

Wang, Hanchen (2021). “Stock Price Prediction Based on

Machine Learning Algorithms.” Modern Industrial IoT,

Big Data and Supply Chain, 111–18.

https://www.kaggle.com/datasets/pranavuikey/black-

friday-sales-eda

Comparison of Enhanced XGBoost Algorithm with Light Gradient Boosting Machine to Determine the Prediction of Black Friday Sales

479