Analysing the Effect of Platform and Operating System Features on

Predicting Consumers’ Purchase Intent using Machine Learning

Algorithms

Ramazan Esmeli

, Alaa Mohasseb

and Mohamed Bader-El-Den

School of Computing, University of Portsmouth, Portsmouth, U.K.

Keywords:

Purchase Intention Prediction, Purchase Behaviour Prediction, Browsing Behaviour, Classiﬁcation, Machine

Learning.

Abstract:

Predicting future consumer browsing and purchase behaviour has become crucial to many marketing plat-

forms. Consumer purchase intention is one of the main inputs used as a measurement for consumer demand

for new products. In addition, identifying consumers’ purchase intent play an important role in recommender

systems. In this paper, the effect of using different platforms on users’ behaviours is explored. In addition, the

effect of users’ platforms and their purchase intentions behaviours are investigated. We conduct computational

experiments using different machine learning algorithms in order to investigate the using users’ operating sys-

tem and platform types as features. The results showed that the users’ purchase intentions and behaviours are

correlated with these features.

1 INTRODUCTION

Predicting future consumer browsing and purchase

behaviour has become crucial to many marking plat-

forms. Knowing customers need plays an impor-

tant role in identifying potential customers (Christy

et al., 2018). Moreover, prior information and expert

knowledge is an important aspect to obtain reliable

and consistent solutions (Mohasseb et al., 2019).

Consumer purchase intention is one of the main

inputs that marketing managers use to forecast fu-

ture sales and to determine how the actions they take

will impact consumers’ purchasing behaviour (Mor-

witz et al., 2014),(Esmeli et al., 2020). In addition,

these intentions are used as a measurement for con-

sumer demand of new products, in which marketing

managers could use as an in indicator of future de-

mand for their products, and to assess how their mar-

keting actions will impact that future.

Furthermore, identifying consumers’ purchase in-

tent play an important role in recommender systems

when determining which users will be receptive to

speciﬁc product recommendations (Korpusik et al.,

2016), (Esmeli et al., 2019).

https://orcid.org/0000-0002-2634-6224

https://orcid.org/0000-0003-2671-2199

https://orcid.org/0000-0002-1123-6823

The most common approach taken by many recent

studies is to identify the next buyer or the buyer in-

tent using machine learning algorithms such as deep

learning (Salehinejad and Rahnamayan, 2016), (Ko-

rpusik et al., 2016), other works such as (Kim et al.,

2003) used multiple classiﬁer while authors in (Qiu

et al., 2015) and (Raphaeli et al., 2017) used unsuper-

vised learning such as association rules. Moreover,

consumers buying behaviour is analysed using differ-

ent methods such as statistical methods (Gupta and

Pathak, 2014) and hidden Markov model (Norouzi

and Alizadeh, 2016) to identify and recommend a

given product or service (Mart

ınez et al., 2020),

(Raphaeli et al., 2017), (Bang et al., 2013) (Wang

et al., 2015), (Kaatz et al., 2019) and (Liu et al., 2019).

In this paper, the effect of using different plat-

forms and operating systems that users browse web-

sites on users’ behaviours, is explored. In order to

achieve this aim, the following objectives are deﬁned:

1. Analysing customers buying behaviour and their

intent when using different operating systems.

2. Creating a framework to analyse the purchase pre-

diction performance when different operating sys-

tems and platform types are utilised as features.

3. Investigating customers buying behaviour when

different machine learning algorithms are used.

4. Evaluating the performance of different machine

Esmeli, R., Mohasseb, A. and Bader-El-Den, M.

Analysing the Effect of Platform and Operating System Features on Predicting Consumers’ Purchase Intent using Machine Learning Algorithms.

DOI: 10.5220/0010176803330340

In Proceedings of the 12th International Joint Conference on Knowledge Discovery, Knowledge Engineering and Knowledge Management (IC3K 2020) - Volume 1: KDIR, pages 333-340

ISBN: 978-989-758-474-9

333

learning algorithms on the prediction and the clas-

siﬁcation of customers intent when using platform

types and operating systems users are using to

connect to different websites.

The paper is organised as follows: In Section 2,

we discuss related work on consumer behaviour

analysing and purchase prediction. Section 3 de-

scribes the details of the datasets used in our proposed

method. In Section 4, we give experimental results

and discuss the performance of the proposed meth-

ods. Finally, in Section 5, we conclude the paper and

give a new research direction.

2 RELATED WORKS

In this section, we review the existing literature on

consumer purchase prediction using different meth-

ods. Recent studies used machine learning algo-

rithms. In (Kim et al., 2003), authors proposed a Ge-

netic Algorithm (GA)-based multiple classiﬁers com-

bining method for the prediction of the customers’

purchase behaviour. The proposed approach com-

bines individual decisions by multiple classiﬁers. The

experiment for a case from a leading Ecommerce

company in Korea showed that the proposed com-

bined method outperforms any individual classiﬁers.

Furthermore, two experiments were conducted. The

ﬁrst experiment showed that the proposed algorithm

could improve the prediction accuracy of the purchase

propensity and the second experiment, showed that

this method has better performance than other com-

bining methods. Works in (Korpusik et al., 2016) used

deep learning techniques for predicting customer pur-

chase behaviour from Twitter data that recommender

systems could leverage. Authors collected a labelled

corpus of buy/not buy users and their tweets. A neu-

ral network-based classiﬁer performed best at pre-

dicting whether a tweet is relevant to purchase be-

haviour, with an accuracy of 81.2% on mobile de-

vices and 80.4% on cameras. Results showed that

the use of a deep learning model that incorporates

sequential information performed better than ignor-

ing sequential information for the purchase prediction

task. In (Gupta and Pathak, 2014) authors proposed

a framework which combined different techniques of

machine learning, data mining and statistical methods

to predict the purchase behaviour of an online cus-

tomer by selecting an appropriate price range based

on dynamic pricing. The proposed framework has

been tested on a large dataset for an e-commerce ﬁrm.

While (Salehinejad and Rahnamayan, 2016) proposed

a customer behaviour prediction model using recur-

rent neural networks (RNNs) based on the client loy-

alty number (CLN), recency, frequency, and monetary

(RFM) variables. The experiment results showed that

RNNs could predict RFM values of customers efﬁ-

ciently. This model can be later used in recommender

systems for exclusive promotional offers and loyalty

programs management.

Moreover, other studies used unsupervised learn-

ing, such as association rules. Authors in (Qiu

et al., 2015) proposed a predictive framework for cus-

tomer purchase behaviour in the e-commerce con-

text called CustOmer purchase pREdiction modeL

(COREL). Associations among products were inves-

tigated and exploited to predicate customer’s motiva-

tions, and customer preferences for product features

were learned and subsequently used to identify the

candidate products most likely to be purchased. In

addition, in this study, the authors investigated three

categories of product features and develop methods

to detect customer preferences for each of these three

categories. Experiments conducted on a real dataset

showed that customer preference for particular prod-

uct features plays a key role in decision-making and

that COREL greatly outperforms the baseline meth-

ods. While in (Raphaeli et al., 2017) authors used

clickstream to compare browsing behaviour in mo-

bile and PC sessions. Analysis has been conducted

to identify the differences in site usage characteris-

tics across different channels, in which the analysis

showed that user engagement in pc session is higher

than mobile sessions. In addition, sequential asso-

ciation rule mining was applied to the clickstream

data navigation patterns, which represent formal com-

binations of web page transitions. The ﬁnal analy-

sis demonstrated that browsing behaviour tends to be

more task-oriented in mobile sessions compared to

PC sessions.

Furthermore, consumers buying behaviour is anal-

ysed using different methods such as graphical prob-

ability model. Works in (Wen et al., 2018) proposed

a graphical probability model that exploits the pay-

ment data to discover customer purchase behaviour

in the spatial, temporal, payment amount and product

category aspects, named STPC-PGM. As a result, the

mobility behaviour of an individual user showed that

it could be used to predict with a probabilistic graph-

ical model that accounts for all aspects of each cus-

tomer’s relationship with the payment platform. In or-

der to achieve real-time advertising, an online frame-

work was developed that efﬁciently computes the pre-

diction results. The experiment results showed that

STPC-PGM is effective in discovering customers’

proﬁling features, and outperforms the state-of-the-

art methods in purchase behaviour prediction. While

in (Norouzi and Alizadeh, 2016), a hidden Markov

KDIR 2020 - 12th International Conference on Knowledge Discovery and Information Retrieval

334

model was used, in which authors proposed a hybrid

method based on hidden Markov model and multi-

nomial choice model to analyse customer choice be-

haviour. The proposed method was applied to the re-

tail store data set.

In addition, authors in (Mart

ınez et al., 2020) de-

veloped advanced analytics tools that predict future

customer behaviour in the non-contractual setting. A

dynamic and data-driven framework was built for pre-

dicting whether a customer is going to make a pur-

chase at the company within a certain time frame in

the near future. For that purpose, the authors pro-

posed a new set of customer-relevant features that

derive from times and values of previous purchases.

These customer features are updated every month,

and state-of-the-art machine learning algorithms are

applied for purchase prediction. Using a data set

containing more than 10, 000 customers and a total

number of 200, 000 purchases an accuracy score of

89% was obtained and an Area Under Curve(AUC)

value of 0.95 for predicting next month purchases on

the test data set. (Choudhury and Nur, 2019) pro-

posed an engineered approach to classifying potential

customer based on previously recorded purchase be-

haviour. Machine learning algorithms were applied

to ﬁnd a pattern and predict potential customers by

using the classiﬁcation results as ground truth. The

proposed approach achieved an accuracy of 99.4%.

Authors in (Bang et al., 2013) investigated con-

sumer browsing and purchase behaviour. The result

showed that the frequency and purchase time irregu-

larity was found to have a positive impact on mobile

commerce adoption. The results also suggested that

search cost inﬂuences the decision to adopt mobile

commerce. In addition, the consumers who search

multi-item or categories at a time, engaged in ac-

tive search, and conducted a thorough search, are less

likely to adopt mobile-commerce. Authors in (Wang

et al., 2015) used a dataset from an internet-based

grocery retailer to evaluate the changes in customers’

spending behaviour when M-shopping is adapted,

such as using smartphones, tablets and online. Results

showed that the number of orders placed per year in-

creased as customers adopt M-shopping. In (Carre

et al., 2019), the effect of exposing time to product ad-

vertisement on TV is investigated on users’ purchase

behaviours. They used machine learning models in

order to analyse the inﬂuence of purchase behaviour

from the advertisement. Results show that the effect

of advertisement on TV has a minimal inﬂuence fac-

tor on users’ purchase behaviours.

While, in (Kaatz et al., 2019), the different im-

pacts of different devices were examined, such as

desktop computers, tablets, and smartphones on the

success of various marketing channels dependent

on the device. This study demonstrated that desk-

top users should be approached by marketing chan-

nels which induce casual browsing behaviour (e.g.,

newsletter, social media, referrer, SEO). In contrast,

mobile users are more likely to use marketing chan-

nels that do not require an extended information

search, such as SEA or direct visits to familiar stores.

In addition, customer experience is important in im-

proving attribution outcomes (e.g., conversion rates)

by combining clickstream and survey data to under-

stand consumers’ decision processes. Finally, in (Liu

et al., 2019) authors explored whether consumers’

app adoption stimulates additional purchases and how

this change in purchase behaviour differs across cus-

tomers with different levels of spending share for dif-

ferent product categories and customer loyalty. Trans-

actional data from a Chinese online retailer were used.

Results showed that app adopters have higher pur-

chase incidence, buy more frequently, and spend more

in each order than non-adopters. Furthermore, results

suggest that apps are worth investing in despite their

similarity to mobile websites and can induce non-

loyal customers to purchase more and thus potentially

foster these customers’ loyalty.

3 DATASET ANALYSIS

In order to explore the effect of using different plat-

forms on users’ behaviours, six datasets were used

provided by a UK based content personalisation com-

pany. The dataset statistics are shown in Table 1

give information about the number of interactions, the

number of unique items and users. Also, the datasets

are analysed in terms of the statistics about which

platforms and devices e-shoppers are using to visit the

six e-commerce platforms.

Table 1: The statistics about the number of the items, users

and interactions in the dataset.

Dataset #user #item #interactions

dataset 1 148832 3835 551851

dataset 2 361330 33273 768999

dataset 3 478089 20480 456077

dataset 4 301089 9111 789059

dataset 5 215240 17431 519251

dataset 6 266871 34152 800985

Table 2 shows the the number of operating sys-

tems that users use to connect to each website. It can

be seen in Table 2 that in most websites IOS os has su-

periority over others except dataset 1 in which users

Analysing the Effect of Platform and Operating System Features on Predicting Consumers’ Purchase Intent using Machine Learning

Algorithms

335

mostly use Windows os to connect to websites. More-

over, we eliminated operating systems which have

less than 100 connections to the websites. The oper-

ating systems which have been eliminated are black-

berry, Fedora, Debian, FreeBSD, OpenBSD, Win-

dows 2000, Windows Phone, Windows RT, Firefox

OS and Ubuntu.

Table 3 shows the number of interactions comes

from each platform. Except Dataset 1, in all other

datasets the number of mobile users has more than the

number of users from other platforms. Furthermore,

We analyse the dataset in terms of the distribution of

interaction and its outcome over weekday, day of the

month, operating systems and platforms. We select

only Dataset 6 since it has the highest number of the

interactions, also we have limited space to put other

datasets’ analysis.

Figure 1 shows the number of interactions over the

day of the week by the outcome of the interaction for

dataset 6. This Figure shows that Wednesday has the

most number of interactions among the other day of

the week.

Figure 1: User-Website interaction distribution over the

days of the week by outcome of the sessions for Dataset

Figure 2 shows the frequency of the interactions

on the hours of the day by outcome for dataset 6. It

shows that users mostly active after midday.

Figure 2: User-Website interaction distribution over the

hours of the day by outcome of the sessions for Dataset 6.

We also analysed the session outcome and user-

website interactions distribution for each operating

system for dataset 6. As seen in Figure 3, website

visitors mostly use IOS operating system.

Figure 3: Distribution of User-Website interaction for the

operating systems grouped by outcome of the sessions for

Dataset 6.

Finally, we analyse the dataset in terms of the

number of user interactions and platform types (Fig-

ure 4). As seen in Figure 4, most of the visitors use a

mobile platform.

Figure 4: Distribution of User-Website interaction for the

platforms grouped by outcome of the sessions for Dataset 6

4 EXPERIMENTS, RESULTS AND

DISCUSSION

In this section we give the details of the experiments

and results. Figure 5 shows the brief overview of the

way we follow for the experimental evaluation of the

proposed approach.

Three well-known machine learning prediction

models are used to explore the effect of users’ plat-

forms on their purchase intention behaviours. These

prediction models are K-Nearest Neighbour (KNN),

Decision Tree (DT), and Bagging classiﬁer. The rea-

son of selecting three different classiﬁers is that they

have different ways of model learning. For exam-

ple, Bagging classiﬁer is a type of ensemble learn-

ing models (Brzezinski et al., 2018),(Bader-El-Den

KDIR 2020 - 12th International Conference on Knowledge Discovery and Information Retrieval

336

Table 2: Statistics about Operating Systems Users Use to Visit the Websites.

Operating Systems Dataset 1 Dataset 2 Dataset 3 Dataset 4 Dataset 5 Dataset 6

Android 7452 97126 217499 228532 195786 95151

Chrome OS 218 658 8002 3718 1609 3234

Linux 28456 2446 2189 864 19723 2008

Mac OS X 14250 47493 96197 57929 83075 95662

Other 31719 77735 2483 24025 19764 18437

Windows 21692 4888 7682 1976 6065 16312

Windows 10 141310 106667 103470 109707 163579 131327

Windows 7 66234 58595 28607 22115 87893 30549

Windows 8 674 343 560 523 484 315

Windows 8.1 10953 3073 6435 5613 54067 5394

Windows XP 911 272 500 443 874 270

iOS 12298 475057 370670 431394 210895 450333

Table 3: Statistics about the Platforms users use to connect

to websites.

Dataset Mobile Pc Tablet

Dataset 1 16191 317286 3562

Dataset 2 522675 303614 48467

Dataset 3 542152 259108 45921

Dataset 4 530621 229191 128246

Dataset 5 438576 304657 102158

Dataset 6 307616 298133 244454

et al., 2018). Furthermore, in order to validate the

proposed approach and evaluate the ability of the ma-

chine learning prediction models, we run experiments

with ten cross-validations. The F-score was used as

the metric to measure the performance of the models.

F-score reﬂects both precision and recall metrics (Zhu

et al., 2019) and common metric in order to validate

the performance of prediction models.

4.1 Feature Engineering

In order to train the classiﬁcation models the follow-

ing features were selected:

1. Day of the Week: This feature indicates the day

of the week the user interacted with the platform.

This feature has an important indication of users’

purchase behaviour since most of the users prefer

to buy products on the weekdays

2. Hour of Day: Shows the hour that a user has

browsed the products. Statistical analysis shows

that users mostly buy products after working

hours.

3. Session Duration: Shows how many minutes a

session lasted.

4. Longitude: Longitude parameter of the location

that a user connected from to a website.

5. Latitude: Latitude parameter of the location that a

user connected from to a website.

6. number of previous purchases: shows the number

of the products a user purchased in the previous

sessions.

7. number of previous sessions: indicates the num-

ber of times a user has visited the website.

8. number of pages viewed: shows the number of the

viewed pages in the session.

We have added operating systems and the platform

types as features to analyse their effect on the users’

purchase behaviour prediction.

1. operating systems (os): shows which operating

system is used to browse the website.

2. platform types (ff): shows which platform type

(mobile, tablet, pc) a user is using to browse the

website.

4.2 Results

In this section, the results of the machine learning al-

gorithms are presented and analysed for each of the

six datasets. Table 4 shows the overview of the re-

sults for all of the datasets. In the following sections,

the results will be discussed in more details for each

dataset.

4.2.1 Dataset 1

When comparing the performance of DT, Bagging,

and KNN, DT has a better F-score when the operat-

ing systems (os) and platform types (ff) features are

used. While KNN classiﬁer has performed very sim-

ilar when these attributes are included and excluded.

In addition, KNN classiﬁer shows the worst F-score

Analysing the Effect of Platform and Operating System Features on Predicting Consumers’ Purchase Intent using Machine Learning

Algorithms

337

Figure 5: Experimental design to analyse the effect of using operating systems and platform types as features.

Table 4: Performance of the classiﬁers (K-Nearest Neighbour (KNN), Decision Tree (DT) and Bagging with (+) and without

(-) using operating systems (os) and platform types (ff) as features.

Dataset 1 Dataset 2 Dataset 3 Dataset 4 Dataset 5 Dataset 6

ML + os, ff - os, ff + os, ff - os, ff + os, ff - os, ff + os, ff - os, ff + os, ff - os, ff + os, ff -os, ff

DT 0.593 0.588 0.838 0.835 0.709 0.699 0.828 0.820 0.816 0.806 0.857 0.852

Bagging 0.581 0.574 0.850 0.845 0.699 0.686 0.843 0.832 0.837 0.820 0.874 0.866

KNN 0.395 0.394 0.693 0.694 0.564 0.563 0.631 0.634 0.643 0.644 0.731 0.730

score among the other classiﬁers. Moreover, interest-

ingly, DT classiﬁer has performed better than Bag-

ging. Also, prediction models have the worst F-score

comparing to the other datasets in all classiﬁcation

models (DT:59.3 %, Bagging: 58.1 %, and KNN 39.5

%). Overall, using os and ff types as features has im-

proved the F-score of classiﬁcation models.

4.2.2 Dataset 2

In Dataset 2, Bagging classiﬁer has the best F-score,

followed by DT with 85 % and 83.8 % respectively.

While KNN classiﬁer has the worst F-score. In

this Dataset, classiﬁcation results show that ensemble

classiﬁer has better performance in predicting users’

purchase intention. Also, adding the os and ff types as

features helped to get a better prediction score when

DT and Bagging classiﬁers were used.

4.2.3 Dataset 3

Dataset 3 performance results show the similarity

with the performance results of Dataset 1 in terms of

how the prediction models are performed. When os

and ff were used as features, all the classiﬁers per-

formed better. However, KNN classiﬁer showed al-

most similar performance when os and ff used.

4.2.4 Dataset 4

Bagging classiﬁer outperformed the other classiﬁers

with an 84.3 % when os and ff features are added.

Interestingly, we could not see any improvement on

KNN classiﬁer, while Bagging and DT show better

F-score when os and ff are included.

4.2.5 Dataset 5

Similarly, in Dataset 4, adding os and ff did not help to

improve the performance of the KNN classiﬁer, while

Bagging gives the best F-score with 84.3 % when os

and ff are included as features.

4.2.6 Dataset 6

Classiﬁer models produced the highest performance

scores in this dataset comparing to other datasets.

Also, including os and ff as features helped to get bet-

ter performance in all classiﬁers. In addition, Bagging

has the best F-score, while KNN is the lowest, with

87.4% and 73.1% respectively.

4.2.7 Overall Result Analysis

DT and Bagging show the superiority over the KNN

classiﬁer on all datasets for purchase prediction. On

the other hand, using os and ff as features does not

have the same effect as DT and Bagging for each

dataset. Generally, os and ff have an indication role

over users’ purchase intention and using these fea-

tures is useful to determine purchase prediction. In-

terestingly, KNN classiﬁer did not show the positive

effect when os and ff are added as features to train the

classiﬁcation models on datasets 2,4 and 5.

4.3 Discussion

The effects of adding new features that show the

users’ behaviours are investigated on performance im-

provement for purchase intention prediction (Mokryn

et al., 2019) (Carre

on et al., 2019). Authors in

KDIR 2020 - 12th International Conference on Knowledge Discovery and Information Retrieval

338

(Mokryn et al., 2019) investigated the feature of prod-

uct trendiness on purchase prediction performance.

They found a signiﬁcant difference when the prod-

uct trendiness was included. In addition, works in

(Wang et al., 2015) analysed the correlation between

devices that are connected to the users when visit-

ing an e-commerce website and their purchase be-

haviours. They found that there is a positive corre-

lation in their buying behaviour and the usage of mo-

bile devices to visit e-commerce websites. However,

they did not investigate the effect of using mobile de-

vices on purchase prediction using machine learning

algorithms. In our research, we conducted an exper-

imental analysis in order to analyse the effect of op-

erating systems and platform types on the prediction

of users’ purchase intention using machine learning

algorithms. The results showed that similar to con-

sidering product trendiness as feature (Mokryn et al.,

2019); there is a positive correlation on purchase pre-

diction performance in the sessions when the machine

learning algorithms are aware of users’ operating sys-

tems and platform types. The limitation of this work

is that we compared the difference of the machine

learning algorithms’ performance on purchase predic-

tion before and after adding both features however

one of the features(os or ff) could lead better corre-

lation result.

5 CONCLUSION AND FUTURE

WORKS

In this work, we investigated the effect of the op-

erating system and platform types that users use to

browse the e-commerce platforms on their purchase

behaviour. In order to identify their impact, we

run computational experiments with and without in-

cluding operating system and platform types as fea-

tures. The results showed that the users’ purchase be-

haviours are correlated with these features.

As future research, we will integrate the purchase

prediction results with session-based recommender

systems in order to improve personalisation when

users browse the products in a session.

REFERENCES

Bader-El-Den, M., Teitei, E., and Perry, T. (2018). Biased

random forest for dealing with the class imbalance

problem. IEEE transactions on neural networks and

learning systems, 30(7):2163–2172.

Bang, Y., Han, K., Animesh, A., and Hwang, M. (2013).

From online to mobile: Linking consumers’ online

purchase behaviors with mobile commerce adoption.

In PACIS, page 128.

Brzezinski, D., Stefanowski, J., Susmaga, R., and Szczech,

I. (2018). Visual-based analysis of classiﬁcation mea-

sures and their properties for class imbalanced prob-

lems. Information Sciences, 462:242–261.

Carre

on, E. C. A., Nonaka, H., Hentona, A., and Yamashiro,

H. (2019). Measuring the inﬂuence of mere exposure

effect of tv commercial adverts on purchase behavior

based on machine learning prediction models. Infor-

mation Processing & Management, 56(4):1339–1355.

Choudhury, A. M. and Nur, K. (2019). A machine learn-

ing approach to identify potential customer based on

purchase behavior. In 2019 International Conference

on Robotics, Electrical and Signal Processing Tech-

niques (ICREST), pages 242–247. IEEE.

Christy, A. J., Umamakeswari, A., Priyatharsini, L., and

Neyaa, A. (2018). Rfm ranking–an effective approach

to customer segmentation. Journal of King Saud

University-Computer and Information Sciences.

Esmeli, R., Bader-El-Den, M., and Abdullahi, H. (2020).

Using word2vec recommendation for improved pur-

chase prediction. In IEEE World Congress on Compu-

tational Intelligence (WCCI) 2020. Institute of Elec-

trical and Electronics Engineers.

Esmeli, R., Bader-El-Den, M., and Mohasseb, A. (2019).

Context and short term user intention aware hybrid

session based recommendation system. In 2019 IEEE

International Symposium on INnovations in Intelli-

gent SysTems and Applications (INISTA), pages 1–6.

IEEE.

Gupta, R. and Pathak, C. (2014). A machine learning frame-

work for predicting purchase by online customers

based on dynamic pricing. Procedia Computer Sci-

ence, 36:599–605.

Kaatz, C., Brock, C., and Figura, L. (2019). Are you

still online or are you already mobile?–predicting the

path to successful conversions across different de-

vices. Journal of Retailing and Consumer Services,

50:10–21.

Kim, E., Kim, W., and Lee, Y. (2003). Combination of mul-

tiple classiﬁers for the customer’s purchase behavior

prediction. Decision Support Systems, 34(2):167–175.

Korpusik, M., Sakaki, S., Chen, F., and Chen, Y.-Y. (2016).

Recurrent neural networks for customer purchase pre-

diction on twitter. In CBRecSys@ RecSys, pages 47–

50.

Liu, H., Lobschat, L., Verhoef, P. C., and Zhao, H. (2019).

App adoption: The effect on purchasing of customers

who have used a mobile website previously. Journal

of Interactive Marketing, 47:16–34.

Mart

ınez, A., Schmuck, C., Pereverzyev Jr, S., Pirker, C.,

and Haltmeier, M. (2020). A machine learning frame-

work for customer purchase prediction in the non-

contractual setting. European Journal of Operational

Research, 281(3):588–596.

Mohasseb, A., Bader-El-Den, M., and Cocea, M. (2019). A

customised grammar framework for query classiﬁca-

tion. Expert Systems with Applications, 135:164–180.

Analysing the Effect of Platform and Operating System Features on Predicting Consumers’ Purchase Intent using Machine Learning

Algorithms

339

Mokryn, O., Bogina, V., and Kuﬂik, T. (2019). Will this ses-

sion end with a purchase? inferring current purchase

intent of anonymous visitors. Electronic Commerce

Research And Applications, 34:100836.

Morwitz, V. et al. (2014). Consumers’ purchase intentions

and their behavior. Foundations and Trends® in Mar-

keting, 7(3):181–230.

Norouzi, P. and Alizadeh, S. H. (2016). An extension of

multinomial choice model for customer purchase be-

havior analysis. In 2016 Artiﬁcial Intelligence and

Robotics (IRANOPEN), pages 61–66. IEEE.

Qiu, J., Lin, Z., and Li, Y. (2015). Predicting customer pur-

chase behavior in the e-commerce context. Electronic

commerce research, 15(4):427–452.

Raphaeli, O., Goldstein, A., and Fink, L. (2017). Analyzing

online consumer behavior in mobile and pc devices:

A novel web usage mining approach. Electronic com-

merce research and applications, 26:1–12.

Salehinejad, H. and Rahnamayan, S. (2016). Customer

shopping pattern prediction: A recurrent neural net-

work approach. In 2016 IEEE Symposium Series on

Computational Intelligence (SSCI), pages 1–6. IEEE.

Wang, R. J.-H., Malthouse, E. C., and Krishnamurthi, L.

(2015). On the go: How mobile shopping affects

customer purchase behavior. Journal of Retailing,

91(2):217–234.

Wen, Y.-T., Yeh, P.-W., Tsai, T.-H., Peng, W.-C., and Shuai,

H.-H. (2018). Customer purchase behavior prediction

from payment datasets. In Proceedings of the Eleventh

ACM International Conference on Web Search and

Data Mining, pages 628–636.

Zhu, G., Wu, Z., Wang, Y., Cao, S., and Cao, J. (2019).

Online purchase decisions for tourism e-commerce.

Electronic Commerce Research and Applications,

38:100887.

KDIR 2020 - 12th International Conference on Knowledge Discovery and Information Retrieval

340