See discussions, stats, and author profiles for this publication at: https://www.researchgate.

net/publication/384243694

FORECASTING OF SUPERMARKET SALES USING BIG DATA ANALYTICS AND

MACHINE LEARNING TECHNIQUES IN BUSINESS SECTOR

Article · April 2023

DOI: 10.5281/zenodo.13932366

CITATIONS READS

3 324

1 author:

Davinder Pal Singh

Salesforce.com
7 PUBLICATIONS 7 CITATIONS

SEE PROFILE

All content following this page was uploaded by Davinder Pal Singh on 23 September 2024.

The user has requested enhancement of the downloaded file.

 International Journal of Core Engineering & Management
Volume-7, Issue-06, 2023 ISSN No: 2348-9510

FORECASTING OF SUPERMARKET SALES USING BIG DATA ANALYTICS AND

MACHINE LEARNING TECHNIQUES IN BUSINESS SECTOR

Davinder Pal Singh

Technical Architect
Salesforce
Toronto, Canada
[email protected]

Abstract

In the modern digital age, the conventional approach to business analysis has been altered as a
result of advancements in machine learning. As marketplaces evolve, customers and businesses
need to properly predict future markets and behaviour to achieve sustainable success. These
advanced technologies have revolutionised how organisations carry out data analysis, gain
knowledge, and come to effective decisions. One of the recent trends is a popularity of predictive
analytics as part of business analytics. Other related abilities that several of the algorithms share
include the capacity to analyse vast databases of earlier occurrences in requests to identify
patterns and trends, which aid an organisation in generating accurate prognoses relating to future
activities. The use of big data analytics methods to enhance retail sales forecasting has gained
popularity in the last several years. Using big data analytics, this article compares and contrasts
several ML methods for forecasting sales at supermarkets. The 2013 BigMart Sales dataset is
utilised in this study, which explores the use of ML algorithms to predict retail sales patterns. A
variety of thorough preprocessing techniques were used, such as PCA feature extraction, outlier
detection, and handling of missing variables. F1-score, recall, accuracy, and precision metrics were
utilised to assess a variety of classification models, including XGBoost, GLM, Decision Tree, and
KNN. With an accuracy of 84.7%, the results show that KNN performed best, demonstrating its
potency in forecasting sales patterns.
Index Terms—Component, Data Analytics, Sales prediction, Machine learning, KNN, Decision
tree, XGBoost, Generalized Linear Model, Dimensionality reduction.

I. INTRODUCTION
Various types of shopping centres, including supermarkets and retail shops, keep records of the
products and things sold, including information on the customers and their dependent and
independent traits and attributes[1][2]. These records also include information about certain assets.
Every market tries to give personalised and limited-time deals in an effort to attract more
customers in a short amount of time. Thus, the collected data may potentially be used to predict
future sales using ML algorithms. Numerous problems face an organisation when it lacks an
accurate sales forecast model. Retailers, distributors, and manufacturers may all benefit from sales
forecasts. Long-term projections facilitate business expansion, while short-term forecasts help with
inventory control and production scheduling. In industries where goods have a limited shelf life,
sales forecasting is essential to reduce revenue loss during periods of excess and scarcity [3][4].
18
 International Journal of Core Engineering & Management
Volume-7, Issue-06, 2023 ISSN No: 2348-9510

To assist supermarkets, make data-driven choices and optimise their operations, data analytics is
used in supermarkets to analyse customer data and forecast future sales patterns. Supermarkets
may generate targeted marketing strategies that improve customer loyalty and boost sales income
by analysing consumer data to target certain customer groupings[5][6].
A growing number of businesses are turning to big data analytics to make sense of the mountain
of data that is available from places like social media, customer loyalty programmes, and point-of-
sale systems[7]. Supermarkets may use big data analytics to examine this information and learn
about consumer trends, preferences, and behaviour. Supermarkets may find useful insights,
patterns, and sales trends for the future by applying ML algorithms to the data and creating
predictive models[8].
Business and industry trends may also be analysed on a time scale with the use of data mining and
big data. It also brings together statistical approaches, programming processes, ML algorithms,
and data engineering. AI calls for and greatly benefits from solid background in mathematics,
statistics, information science, and computer science. A branch of "Computer Science" (or
"Artificial Intelligence"), "machine learning" is the study of algorithms derived from stochastic
theory that can effectively carry out tasks in the absence of explicit programme instructions by
drawing conclusions and capitalising on patterns [9][10]. The mathematical model of the data used
for training purposes, sometimes referred to as "Training Data," is constructed using ML
techniques in order to extract conclusions and assumptions from the data [11]. Supermarket sales
datasets may be used as a springboard for new insights into supervised and unsupervised issue
types in ML, with the former often serving as a source for classification-type challenges[12].
The motivation behind this study is to address the important problem of precisely forecasting sales
trends in retail settings, with a focus on the extensive 2013 BigMart Sales dataset. For supermarkets
and other retail establishments, accurate sales forecasting is essential to strategic planning,
inventory control, and overall operational efficiency. With the use of cutting-edge machine
learning methods like K-Nearest Neighbours (KNN), Generalized Linear Models (GLM), Decision
Trees, and XGBoost, this study attempts to create reliable predictive models that can manage
intricate sales dynamics. The goal of this research is to offer practical implications that would help
retail decision-makers increase the retail industry‟s profitability and customer satisfaction in the
current highly competitive environment. This is achieved through rigorous preprocessing, feature
extraction, and model evaluation employing metrics like recall, F1-score, accuracy, and precision.
The key contributions of the study on forecasting sales using the BigMart Sales dataset:
 This study introduces a comprehensive preprocessing including improved handling of
missing values through average weight imputation and mode-based filling for outlet size,
along with outlier detection and removal, ensured a cleaner dataset for analysis.
 Utilisation of feature extraction techniques such as PCA for dimensionality reduction and
creation of new features helped in capturing essential information and optimising model
performance.
 Comprehensive evaluation of multiple classification models (XGBoost, GLM, Decision
Tree, KNN) provided insights into their effectiveness for sales prediction, highlighting
KNN as the top performer.
 Visual representations like correlation matrices and confusion matrices offered clear
insights into relationships between variables and model performance, aiding in better
understanding and interpretation of results.

19
 International Journal of Core Engineering & Management
Volume-7, Issue-06, 2023 ISSN No: 2348-9510

 Rigorous evaluation employing metrics like F1-score, recall, accuracy, and precision
provided a robust assessment of model performance, enabling informed decision-making
for deployment in real-world scenarios.

The paper is organised as follows. Section 2, present a background study of sales prediction using
BigMart sales dataset, Section 3 offers methods and methodology, and Section 4 results analysis
and discussion. Conclusion and future work of this study present in section 5.

II. LITERATURE REVIEW

In this section, provide some previous work on big data analytics based on machine learning. The
field of sale forecasting using ML has seen a lot of work presented so far. An overview of relevant
sales forecasting studies is given in this section. Table 1 shows the results of a comparison study
using ML and big data analytics in the corporate world.
In, Niu, (2020) approach may efficiently mine properties across several dimensions to provide
accurate predictions. The XGBoost sale prediction model is tested in this work using datasets given
by the Kaggle competition and sales data from Walmart stores. The experimental findings
demonstrate that the strategy outperforms the other ML algorithms. This paper's RMSSE measure
is 0.141 times lower than the LR method and 0.113 times lower than the Ridge technique[13].
In, Jiang, Ruan and Sun, (2021) investigates the practicability of several models for Walmart sales
forecasting, including ML-only, hybrid, and conventional time series models. Predictions and
empirical analyses should be based on data collected between 2016-06-19 and 2016-08-14. The
Prophet model and the ML model, lightGBM model, are used for training and testing. The Prophet
model breaks down trends, seasons, and holidays. Sales figures are derived from Walmart
supermarkets and cover the period from 2011-01-29 to 2016-06-19. The results demonstrate that the
ML model accurately forecasts sales at retail stores; the LightLGB model and the Prophet model
both have RMSEs of 0.617 and 0.694, respectively[14].
In, Sun et al., (2021) Lines are drawn between the spots in accordance with the time sequence.
Second, for every route map, they determine its fractal dimension using the box-counting
approach. Lastly, in order to enhance the KNN approach, they took the similarity of distributions
and Gaussian functions into account. Our upgraded KNN customer categorisation model
outperforms SVM and classic KNN models with a better F1-score of 0.926 and an accuracy greater
at 0.925[15].
In Zhao et al. (2021) categorise customers according to their purchase history: high volume, low
volume, and no volume. This is done using point-of-sale data. Using an LSTM-NN, the experiment
establishes a model for consumer behaviour categorisation. When compared to LR and SVM,
LSTM-NN achieves a greater recognition accuracy of 5.26% and 6.97%, respectively, according to
numerical testing[16].

20
 International Journal of Core Engineering & Management
Volume-7, Issue-06, 2023 ISSN No: 2348-9510

TABLE I. COMPARATIVE STUDY ON BIG DATA ANALYTICS AND MACHINE LEARNING

TECHNIQUES IN BUSINESS SECTOR
References Methodology Dataset Performance Limitations & Future Work
[5] Recurrent Sales data of 800 RMSE: 0.039 Further tuning of RNN
Neural products over 49 parameters; exploration of other
Network weeks neural network architectures.
(RNN)
[6] XGBoost Walmart RMSSE: 0.141 Exploration of additional
supermarkets sales (Logistic features; comparison with other
data from Kaggle Regression), 0.113 advanced machine learning
competition (Ridge) models.
[7] Prophet Walmart RMSE: 0.694 Combination of more models for
model and supermarket sales (Prophet), 0.617 better performance; further
LightGBM data (LightGBM) validation on different datasets.
[8] Improved Customer Accuracy: 0.925, Application to larger and more
KNN using classification based F1-score: 0.926 diverse datasets; integration with
fractal on fractal dimension other classification techniques.
dimension and stay time
[9] Long Short- POS data with Improved Refinement of the LSTM model;
Term consumer behaviour accuracy: 5.26% exploration of additional
Memory variables (Logistic behavioural and contextual
Neural Regression), 6.97% variables for classification.
Network (SVM)

A. Research gaps
Despite advancements in various sales forecasting models like RNNs, XGBoost, Prophet,
LightGBM, and improved KNN algorithms, a significant research gap exists in integrating these
models to create hybrid approaches for more robust predictions. Current studies often focus on
individual models or comparisons, lacking exploration into combined models that leverage
strengths and mitigate weaknesses. Moreover, there is limited research on the interpretability of
these models, crucial for practical applications in retail and marketing. Additionally, while studies
use datasets from Walmart and market products, more diverse datasets across various market
segments and regions are needed to validate model generalizability. Closing these gaps could
enhance reliability and applicability in sales forecasting.

III. MATERIALS AND METHODS

The proposed system aims, among other things, to maximise revenue by locating a dependable
method for predicting sales trends via the use of ML. In order to accomplish the organisation's
goal, these models may be utilised in many domains and taught to meet expectations. Using a 2013
BigMart Sales dataset, the work's goal is to forecast sales. As a result, the given dataset was split
into subgroups for fitting and validation for enhancing analysis efficiency. During data
preprocessing, to handle missing values in the „Item Weight column, the average weight of the
item was considered to be implanted for the absent values while to fill the missing values in the
„Outlet Size‟ column, the mode according to the outlet type was considered. To further
improvement of the model, outliers were identified and later corrected. Some measures were taken
21
 International Journal of Core Engineering & Management
Volume-7, Issue-06, 2023 ISSN No: 2348-9510

to expand the quality of the dataset, such as feature creation and feature reduction through the
help of PCA. Following that, the data was split into a testing set and a training set. For sales
prediction classification models applied include XGBoost, KNN, Generalized Linear Model (GLM),
Decision Tree, etc. These models were evaluated using recall, accuracy, precision and F1-score. The
comparison investigation shed light on how well various BigMart sales forecast methods
performed.
The proposed research recommended the following procedures for predicting sales of different
categories employing a retail store's sales data. A proposed system's architectural diagram is
shown in Figure 1. This is a detailed overview of all the processes involved.

Big Mart Data

Dataset Collection

Data Preprocessing

Removing Outliers

Feature Missing Values

Extraction

Data Splitting

Training Testing

Classification Models
 XGBoost
 GLM
 DT
 KNN
Model Evaluation
Accuracy, Precision,
Recall, F1-Score

Results

Fig. 1. Flowchart of proposed methodology for BigMart Sales dataset.

A. Data Collection
This comparative study uses of the BigMart Sales dataset, which was gathered in 2013. The input
dataset is divided into 2 subsets and provided valuable insights for analysis. The BigMart Sales
dataset, gathered in 2013 is utilised to predict consumer behaviour. There are two subsets of this
dataset: a test set and a training set. There are 5681 records with 11 attributes in the test set and
8523 records with 12 attributes in the training set, as shown in Table 2. Both independent and
22
 International Journal of Core Engineering & Management
Volume-7, Issue-06, 2023 ISSN No: 2348-9510

dependent variables are present in the training set. The attributes are described in full below:

TABLE II. ATTRIBUTES AND VARIABLES OF THE DATASET

Item Identifier Product ID

Item Weight Weight of Product
Item Fat Content Fat content of Product- Low/Regular
Item Visibility Parameter to know the visibility/reach of
product
Item Type Category of Product
Item MRP Maximum Retail Price of the Product.
Outlet Identifier Store ID
Outlet Establishment The Year in which store is established
Year
Outlet Size Area-wise distribution of Stores-
Low/Medium/High
Outlet_Location_Type Type of city in which outlet is located
Outlet Type Type of outlet Grocery store or supermarket
Item Outlet Sales Sale price of product - The dependent
variable to be predicted

Fig. 2. Sale Prediction of over all products

Figure 2 presents the sales forecast for each item in an understandable manner. From the data, it
can be concluded that people are more likely to buy fruits and veggies than other goods.
Moreover, almost similar quantities of snack food are bought.

23
 International Journal of Core Engineering & Management
Volume-7, Issue-06, 2023 ISSN No: 2348-9510

Fig. 3. Sale depends on Item MRP

Figure 3 shows the scatter plot of "Item_MRP" (Maximum Retail Price) and "Item_Outlet_Sales."
Each point is an observation, with the x-axis displaying Item_MRP and the y-axis showing
Item_Outlet_Sales. The plot illustrates that Item_Outlet_Sales rise with Item_MRP. Four vertical
bands cluster the data points, suggesting four Item_MRP ranges or categories. Higher Item_MRP
increases point density and band spread, indicating a wider sales fluctuation for higher-priced
items.

B. Data Preprocessing
The acquired data underwent preprocessing and cleaning to eradicate any potential flaws, such as
missing values or outliers, which might impact the accuracy of the ML models. Preprocessing
refers to the process of cleaning a dataset of any excessive or irrelevant data. This step often
handles outliers in the dataset and imputes missing values. The dataset has no information for the
columns labelled "Outlet Size" and "Item Weight." Item weight may be thought of as a numerical
variable, while outlet size is categorical. To fill in the blanks when data are absent, we take the
sample weight as a whole and impute it as Item Weight. Since Outlet Size is not a continuous
variable, we must rely on the mode approach to fill in the blanks since we cannot compute the
average. Therefore, by figuring out the size mode based on outlet type, the missing digits in the
outlet size may be discovered.
 Missing Values: The absence of a single data point for a particular variable in a dataset is
known as a missing value. Many other symbols, such "NA" or "unknown," or empty cells
may stand in for them. The absence of these data points makes data analysis more difficult
and increases the risk of biassed or incorrect conclusions.
 Removing Outliers: Datasets may sometimes include out-of-the-ordinary, out-of-range
numbers that stand in stark contrast to the rest of the data. Identifying and eliminating
abnormal values, which are known as outliers, may often lead to better model skill and
machine learning modelling in general.

C. Feature Extraction
Feature extraction involves sorting all data into categories in order to extract the most important
and relevant information [17]. It is critical to get all the required information or minimise the loss
of pertinent data while dealing with a big dataset. The data loss rate may be reduced by the use of
24
 International Journal of Core Engineering & Management
Volume-7, Issue-06, 2023 ISSN No: 2348-9510

feature extraction, which helps manage the vital information out of enormous raw datasets.
 Creating new features: A vital step in improving the efficiency of ML algorithms by
obtaining relevant data is the process of extracting new features from old ones.
 Dimensionality reduction: To reduce the number of features while preserving critical
information, one may use approaches like PCA or feature selection.

D. Data Splitting
Data separation is a common step for training and testing the model. In this study, dataset is split
into two parts. The two parts are training and testing.

E. Machine learning Classification Models

For the comparative analysis some classification models for BigMart sales dataset are described in
this section.
1). XGBoost Model
XGBoost is a prominent ML algorithm that consistently ranks among the top in terms of accuracy.
It is widely used for both regression and classification prediction tasks. In ML, it's an application of
gradient-boosted DT made for speed and performance.
2). Generalised Linear Model
Data sets may be used for prediction using the generalised linear model (GLM). An enlarged
version of a linear regression equation is one example of a general linear model. The GLM is an
enhanced version of linear regression that establishes a connection between the model and the
response variable via a link function. As a result, the variance of every measurement is determined
by its predicted value.

(1)

An intelligent stepwise GLM of a dataset array (tbl) is built using a constant model as a starting
point, and predictors are added or removed using stepwise regression. The final variable of the
table is used as the response variable by stepwise. Stepwise uses both forward and backward
stepwise regression to get a final model.
3). Decision Tree
The goal of decision tree learning is to build a DT that represents f or a near approximation of it
from a collection of (x, f(x)) pairings. While it is theoretically possible for the set of pairs to be
exhaustive when domain x is finite, in practice, sets are typically samples from domain X that
could be limitless. If that's the case, one possibility is to seek for a tree that approximates f
throughout the whole domain, instead of just on the data set.
4). K-Nearest Neighbour
Unsupervised ML algorithms include the KNN method. Simply said, it's the algorithm that uses
the similarity principle to assign a label from a predetermined set of labels to an unlabelled item,
or it sorts the new data point into one of the preexisting categories. To illustrate the point, the k-
NN approach may be used to train a model that can distinguish between square and circular
images. In the event that you provide it with an unclassified image, it will automatically assign it
to the square or circle class.
The class of the new data point may be determined by taking a majority vote among surrounding

25
 International Journal of Core Engineering & Management
Volume-7, Issue-06, 2023 ISSN No: 2348-9510

data points. Choosing how many neighbours to use for categorisation is a manual process. In our
study, we have used the Euclidean distance to assess the similarity. Equation (2) provides the
formula for calculating the Euclidean distance[18]:

(2)

A new data point is assigned to one of the predefined classes by a majority vote of its KNN, chosen
according to the Euclidean distance computed using the aforementioned equation.

IV. RESULTS AND DISCUSSION

This section presents the outcomes obtained through the evaluation of a dataset utilised in this
research, including the outcomes, description of a dataset, performance metrics, and classifier
statistics. For evaluating the performance of each model, a total of four various performance
metrics has been used: F1-score, recall, accuracy, and precision. These parameters are given below:
Accuracy
A measure of accuracy is determined by dividing the total number of correct forecasts by the total
number of predicted values, which includes the true predictions themselves. The corresponding
equation (3) is shown below:

(3)

The findings were analysed using well-respected academic performance metrics that centre on the
confusion matrix. The matrix's visual is shown in Figure 4. The four main features of the matrix
display the outcome data, while the matrix itself is an amalgamation of categorisation results. A
true positive (TP) result is one in which the actual value matches the anticipated value of the
classification. True negative (TN) principles are similar, only they centre on zero. The outcome is
referred to as a false positive (FP) when the prediction is 1 and the real value is 0, while the
converse is termed a false negative (FN).
The experiment results of the machine and machine learning models for BigMart sales dataset are
provided in this section. Graphs, tables, and figures make up the findings.

26
 International Journal of Core Engineering & Management
Volume-7, Issue-06, 2023 ISSN No: 2348-9510

Fig. 4. Correlation Matrix of K-NN on sale prediction

The above Figure 5 demonstrates that Item_MRP has a large positive correlation with Sale Price in
the Correlation Matrix, whereas Item Weight and Item Visibility have weaker associations. Item
Weight has a slight positive correlation while Item Visibility is negatively correlated.

Fig. 5. Confusion Matrix

Figure 6 shows the confusion matrix of correlation between Item Weight, Item Visibility,
Item_MRP, and `Item_Outlet_Sales displayed in this heatmap. There is a somewhat favourable
association (0.57) among Item_MRP and Item_Outlet_Sales. There is little link among Item Weight
and Item Visibility and other attributes. The correlation strength is shown by the colour intensity.
The following Table 3 shows a comparison among various machine learning models for
comparative analysis and BigMart sales in terms of performance metrics.

27
 International Journal of Core Engineering & Management
Volume-7, Issue-06, 2023 ISSN No: 2348-9510

TABLE III. COMPARISON BETWEEN VARIOUS MODEL FOR BIGMART SALE PREDICTION
Models Accuracy
XGBoost [19] 61.14
GLM[20] 56.03
Decision Tree [21] 62.0
KNN 84.7

Fig. 6. Bar Graph for Accuracy comparison of models

The bar graph in Figure 7 illustrates the accuracy comparison between various models used for
BigMart sale prediction. Among the models, KNN demonstrates the highest accuracy at 84.7%,
significantly outperforming the other models. This suggests that KNN is particularly effective for
this dataset, possibly due to its ability to capture complex relationships in the data through its
instance-based learning approach. In contrast, XGBoost shows the lowest accuracy at 61.14%,
indicating that, despite its reputation for strong performance in many scenarios, it may not be as
well-suited for this particular task. The GLM and Decision Tree models have moderate accuracy
scores of 56.03% and 62.0% respectively, highlighting a substantial gap between these traditional
models and KNN in terms of predictive accuracy for BigMart sales.

V. CONCLUSION AND FUTURE SCOPE

This project provides an understanding of ML and its critical ideas alongside the data processing
and modelling techniques used in this field. Forecasting of sale at these different stores of Big Mart
by using these methods is the focus here. Since the dataset gathered from BigMart Sales included
extensive records, refining processes strengthened data credibility necessary for prediction. KNN
proved to be the best performing model and therefore indicated that KNN could be used to predict
sales trends with very minimal error margin of 84. 7%. The study brings out the importance of
improving accuracy in forecasting in determining the position of inventory as well as efficiency in
retail operations. Continuing the research of hybrid model approaches and improving
interpretability could lead to improving the predictive performance in the short-term and
achieving strategic agility for retailers to predict and plan resource allocation and targeted tactics
to improve customer satisfaction in response to changes in the market environment.
The usage of such algorithms such as DL and Transfer Learning can be most probably seen in near
28
 International Journal of Core Engineering & Management
Volume-7, Issue-06, 2023 ISSN No: 2348-9510

future. The future work includes filling the gaps that have been pointed out in the study, for
instance, using the combination of the early and late stages of machine learning models. Thus,
there remains potential for further research on bringing interpretability methods to these models,
and these findings can benefit retail and marketing applications. Extension of these models to
other markets and to other regions would also prove the scope of generalisation of these models to
increase reliability about sales.

REFERENCES
1. A. H. Ali, M. Z. Abdullah, S. N. Abdul-Wahab, and M. Alsajri, “A Brief Review of Big Data
Analytics Based on Machine Learning,” Iraqi J. Comput. Sci. Math., 2020, doi:
10.52866/ijcsm.2020.01.02.002.
2. S. J. Isabella and S. Srinivasan, “An understanding of machine learning techniques in big
data analytics: A survey,” Int. J. Eng. Technol., 2018, doi: 10.14419/ijet.v7i3.12.16450.
3. R. Odegua, “Applied Machine Learning for Supermarket Sales Prediction,” no. January,
2022.
4. J. Thomas, “The Effect and Challenges of the Internet of Things (IoT) on the Management of
Supply Chains,” Int. J. Res. Anal. Rev., vol. 8, no. 3, pp. 874–878, 2021.
5. V. Rohilla, S. Chakraborty, and R. Kumar, “Car Auomation Simulator Using Machine
Learning,” SSRN Electron. J., 2020, doi: 10.2139/ssrn.3566915.
6. P. Khare, “The Impact of AI on Product Management : A Systematic Review and Future
Trends,” vol. 9, no. 4, 2022.
7. A. Rath, A. Das Gupta, V. Rohilla, A. Balyan, and S. Mann, “Intelligent Smart Waste
Management Using Regression Analysis: An Empirical Study,” in Communications in
Computer and Information Science, 2022. doi: 10.1007/978-3-031-07012-9_12.
8. R. K. Vinita Rohilla, Sudeshna Chakraborty, “Random Forest with harmony search
optimisation for location based advertising,” Int J Innov Technol Explor Eng, vol. 8, no. 9, pp.
1092–1097, 2019.
9. S. Mann, A. Balyan, V. Rohilla, D. Gupta, Z. Gupta, and A. W. Rahmani, “Artificial
Intelligence-based Blockchain Technology for Skin Cancer Investigation Complemented
with Dietary Assessment and Recommendation using Correlation Analysis in Elder
Individuals,” Journal of Food Quality. 2022. doi: 10.1155/2022/3958596.
10. V. Rohilla, S. Chakraborty, and M. Kaur, “Artificial Intelligence and Metaheuristic-Based
Location-Based Advertising,” Sci. Program., 2022, doi: 10.1155/2022/7518823.
11. A. Nallamekala, S. Vanukuri, and O. Prakash, “Data Science and Machine Learning
Approach to Improve Online Grocery Store Sales Performance,” no. May, pp. 227–233,
2022.
12. V. Rohilla, M. Kaur, and S. Chakraborty, “An Empirical Framework for Recommendation-
based Location Services Using Deep Learning,” Eng. Technol. Appl. Sci. Res., 2022, doi:
10.48084/etasr.5126.
13. Y. Niu, “Walmart Sales Forecasting using XGBoost algorithm and Feature engineering,” in
Proceedings - 2020 International Conference on Big Data and Artificial Intelligence and Software
Engineering, ICBASE 2020, 2020. doi: 10.1109/ICBASE51474.2020.00103.
14. H. Jiang, J. Ruan, and J. Sun, “Application of Machine Learning Model and Hybrid Model

29
 International Journal of Core Engineering & Management
Volume-7, Issue-06, 2023 ISSN No: 2348-9510

in Retail Sales Forecast,” in 2021 IEEE 6th International Conference on Big Data Analytics,
ICBDA 2021, 2021. doi: 10.1109/ICBDA51983.2021.9403224.
15. F. Sun, L. Zhao, Y. Zuo, and Y. Kaneko, “Application of Fractal Analysis for Customer
Classification Based on Path Data,” in IEEE International Conference on Data Mining
Workshops, ICDMW, 2021. doi: 10.1109/ICDMW53433.2021.00040.
16. L. Zhao, Y. Zuo, K. Yada, and M. Liu, “Application of Long Short-term Memory Based
Neural Network for Classification of Customer Behavior,” in Conference Proceedings - IEEE
International Conference on Systems, Man and Cybernetics, 2021. doi:
10.1109/SMC52423.2021.9658703.
17. J. R. Vergara and P. A. Estévez, “A review of feature selection methods based on mutual
information,” Neural Computing and Applications. 2014. doi: 10.1007/s00521-013-1368-0.
18. H. Jégou, M. Douze, and C. Schmid, “Product quantisation for nearest neighbor search,”
IEEE Trans. Pattern Anal. Mach. Intell., 2011, doi: 10.1109/TPAMI.2010.57.
19. V. Chitre, “Big Mart Sales Analysis,” Int. J. Innov. Technol. Explor. Eng., 2022, doi:
10.35940/ijitee.c9833.0411522.
20. et al., “Machine Learning Approach for Big-Mart Sales Prediction Framework,” Int. J. Innov.
Technol. Explor. Eng., 2022, doi: 10.35940/ijitee.f9916.0511622.
21. M. April et al., “Supermarket Sales Prediction Using Regression,” Int. J. Adv. Trends Comput.
Sci. Eng., vol. 10, no. 2, pp. 1153–1157, 2021, doi: 10.30534/ijatcse/2021/951022021.

30

View publication stats