Abstract: - The e-retail sector is growing day by day and the competitive environment is getting harder.

Businesses have to compete with their competitors in order to survive. In parallel with the increasing internet

penetration, the trade volume in E-Retail sites is also increasing therefore the data generated on these sites is

enormous. Understanding these data with traditional analysis methods is difficult due to the size problem

mentioned. Difficult to understand data causes loss of time, money and customers. In recent years, machine-

learning algorithms have been frequently used to analyse these large-sized data and to use them in decision-

making. This study aimed to perform predictive analysis for the product recommendation system established by

using logistic regression, which is a supervised machine-learning algorithm. In addition, the binary

classification algorithm preferred to predict whether customers make a purchase or not. As a result, the

accuracy degree of the model was 79.73%. This study has the potential to affect the understanding of

Today, physical markets or market places digitized

at a high speed. This digitalization is very important

for businesses. In this case, the customers have been

seen as a marketplace; generate huge amounts of

data on the internet. These generated data can help

businesses make decisions in their supply chain

management processes.

E-Commerce includes transactions in many

consumer. Electronic retailing emerged as a result of

the transfer classical retailing activities to virtual

and online platforms. With the growth in the e-retail

sector and the increasing interest, the competitive

environment is getting more and more challenging.

Figure 1. World E-Commerce Volume (Trillion

E-Retail sites aim to increase their sales in which

the data created by the customers on the web pages

large data size in machine learning increases the

success of the trained model [1], [2], [3], [4], [5].

For this reason, machine-learning algorithms used in

e-commerce web pages can be successful in

understanding customer behaviors, making

appropriate product recommendations or making

sales forecasts.

This study aimed to measure the success of the

machine-learning model based on customer

The remaining of the paper organized as follows.

The literature review presented in section 2. Section

3 presents the used techniques and. Section 4

applies and discusses the approach to logistic

regression method to e-retail company. Finally,

section 5 analysis the applied method. and finally

last section summarizes some conclusions and

discusses potential extensions of the research.

reserachers preferred the logistic regression method

for static representation learning to incremental

learning of embedding dynamic networks[10], [11],

[12], [13]. Some studies used the logistic regression

method to obtain strong rules by addressing the

multi-label classification problem [14], [15], [16].

logistic regression is to estimate the βk+1

parameters.

The maximum likelihood requires finding the

[25], [26], [27], [28]. The goal of model includes all

predictive variables which consists of the predicting

the response variable. Logistic Regression calculates

the probability of success based on the probability

of failure.

As a result of the study, it was the job of

mathematicians and consultants in the past, but with

the effect of increasing technological developments

in changing times, more senior managers and

organizations are trying to use these techniques for

The flowchart of the logistic regression model in

this study is shown in Figure 2.

-Construct probability density likelihood functions

of m samples:

It’s Log likelihood Function:

maximize the function are calculated θ. The

probability of the sample y of 1 is estimated, in

which x is the independent variable vector of the

sample y.

Figure 4. Flow chart of the logistic regression

algorithm

In this section, the logistic regression method

performed with the help of the codes below shown

in Figure 7 again.

Figure 8. The Process of Converting Time stamps to

Familiar Format

The event variable, i.e. events such as views, add-to-

cart, and transactions, are interactions collected over

There are events involving 2,664,312 views, 69,332

adding to carts and 22,457 transactions by 1,407,580

unique visitors(in Fig.10).

Figure 10. Total Views, Add to Cart and Purchases.

When the events are analyzed over the pie chart, it

is seen that 96.7% of the events are viewing, 2.5%

are adding to cart and 0.8% are purchasing (Fig.11).

In total, 11,719 visitors made purchases. The ratio

of visitors purchasing products to the total number

made on the products with the CategoryID of 570.

Figure 13. Number of Users Viewing But Not

Making a Purchase

In Figure 14, 10 of the visitors who made a purchase

known to make a purchase at least once. When the

user's actions are queried with the visitor ID, it is

seen that he displays the product with the ID

number "243566", then adds it to the cart and buys it

Figure 14. The Movements of the First 10 Users

Who Made a Purchase, Visitor ID number 505565

While users are viewing a product, it aimed to show

were classified, a product recommendation system

was established and it was aimed to predict whether

customers would make a purchase in real time.

When we examine the largest companies in the

sector in question, it shown that customer

time, customer satisfaction breeds loyalty. Again,

when we look at the biggest companies in the sector,

people want to feel valuable there rather than seeing

an e-commerce site as just a place for shopping. For

this reason, the importance of understanding your

satisfaction and understand the customer by keeping

the customer in the foreground. Since the aim of the

study is to understand the customer, it is thought

that it contributes to the achievement of the

sustainability goal of the companies.