Product Recommendation Systems using Apriori in the Selection of Shoe

based on Android

Harrizki Arie Pradana

, Laurentinus

, Fransiskus Panca Juniawan

and Dwi Yuny Sylfania

Department of Informatics Engineering, STMIK Atma Luhur, Jl. Jend. Sudirman-Selindung,

Pangkalpinang-Bangka,Indonesia

Keywords:

Recommendation Systems, Apriori, Rules, Consequent, Android.

Abstract:

The development of information technology has given a lot of inﬂuence and changes in human life. One of

the areas of life affected is the ﬁeld of trade. Currently, there are still many people who are looking for shoes

manually by going around looking for shoe stores, and the results are wasting time, energy, and transportation

costs. In addition, by buying manually the buyer does not know the experience of other buyers who have

bought shoes with the same model. In addition, by implementing the recommendation system using the

Apriori algorithm, prospective buyers can ﬁnd out the experience of previous buyers. The advantage of the

Apriori algorithm is simpler, more efﬁcient, and able to handle large amounts of data than other algorithms.

This study uses the waterfall model with four stages in it. The system was tested using the Apriori method to

produce rules derived from the pattern of the combination of two items, rules and consequent. The results of

this study are the successful search for the level of support and trust in shoes so that the shoes obtained are

more accurate using data mining.

1 INTRODUCTION

The development of information technology in the

ﬁeld of communication has at least two rapidly de-

veloping technologies. The ﬁrst is a cellular tele-

phone and the second is an internet networked com-

puter (Kasemin et al., 2016). Mobile phones are no

longer only used for a phone call and short messages

but have developed into multifunctional devices or

what we know better as smartphones.

Business processes in the ﬁeld of trade cannot sep-

arate from the use of smartphones based on Android.

Android is developed based on the Linux kernel sys-

tem so that it categorizes in an open operating system.

Most of the leading cellphone vendors are currently

using Android as an operating system (OS) for their

smartphones. Android has transformed into the most

widely used operating system for smartphones in just

a few years after its appearance (Kurniawati et al., ).

However, user privacy remains a key aspect of every

mobile application (Aceto et al., 2018)(Aceto et al.,

2019).

Shoes, which are one of the human needs, have

increased in use over time, and have become an obli-

gation in various ﬁelds of work, education and a fash-

ion trend which has increased the number of models

and brands of shoes from year to year. The process

of ﬁnding shoes for some people is still doing man-

ually. Manually it means that when you want to buy

new shoes, they need to get around in search of a shoe

store that is about to sell shoes they like. This pro-

cess will cause waste of time and transportation costs

which can reduce if the community is not confused in

ﬁnding a shoe store (Badriyah et al., 2018).

One solution to overcome the above problems is

to make a mobile application based on android shoes.

Android choose because it has many advantages over

platforms (desktop, website) and operating systems

(OS) smartphones (such as iOS, Windows Phone,

Symbian). The beneﬁts of Android include large mar-

ket share, open source, easy to use compared to other

platforms and OS, and easy to use anytime and any-

where.

A mobile application that is equipped with rec-

ommended algorithm will help the user to determine

their choice. There are many algorithms that can be

used to make recommendations, in this case the pur-

chase of shoes. One of them is Apriori. This study

uses Apriori because of its advantages in doing gen-

erates candidate items set and tests if they are frequent

(Kavitha and Selvi, 2016).

Applications that made later can be used by cus-

Pradana, H., Laurentinus, ., Juniawan, F. and Sylfania, D.

Product Recommendation Systems using Apriori in the Selection of Shoe based on Android.

DOI: 10.5220/0009909603110318

In Proceedings of the International Conferences on Information System and Technology (CONRIST 2019), pages 311-318

ISBN: 978-989-758-453-4

 2020 by SCITEPRESS – Science and Technology Publications, Lda. All rights reser ved

311

tomers to check product recommendations, prices,

and shoe details in the shop which can then be a ref-

erence for customers in choosing the desired shoes.

Product recommendation system with Apriori algo-

rithm will add to the application. The product rec-

ommendation system allows applications to display

products that might be liked by users, thereby reduc-

ing product search time. Thus, users can make their

choices according to the trend of shoe users. In ad-

dition, the choice can be accelerated so they can save

their resources.

2 LITERATURE REVIEW

2.1 Recommendation Systems

The recommendation system is an application that

provides and recommends an item in making a de-

cision desired by the user. The implementation of

recommendations in order usually predicts an object,

such as ﬁlm and music recommendations. The sys-

tem runs in two ways, namely by collecting user data

directly and indirectly.

Direct data collection is asking the user to rate

an item. While indirect data collection is by observ-

ing the objects, then it could be seen by users on an

ecommerce web. After the observational data col-

lected, then it is processed using a particular algo-

rithm. After that, the results will return to 10 users as

an item recommendation with the parameters of that

user. The recommendation system is also an alterna-

tive in searching for an item sought by users (Rahar-

jana, 2017).

2.2 Apriori Algorithm

Apriori is part of the association rule method, which

serves to ﬁnd item combinations based on items pur-

chased by customers. Types of association rules in-

clude a priori, generalized rule induction, and hash-

based algorithms. Association rule mining is a data

mining technique to ﬁnd associative rules between a

combination of items, for example, analysis of pur-

chases at a supermarket. With the existence of data

and observations, it can be known some possibilities

for a customer to buy bread together with milk.

By utilizing this condition, self-service owners

can take advantage of these conditions by regulating

the placement of goods or designing marketing pro-

motions (Febriansyah and Samsinar, 2018).

There are three stages to determine frequent pat-

terns (Kavitha and Selvi, 2016), namely:

2.2.1 Generate and Test

The ﬁrst step is to determine the 1-itemset frequent

L1 elements by scanning the database. Then remove

all elements from C that do not meet the minimum

criteria.

2.2.2 Join Step

To reach the element at the next level, Ck joins the

frequencies of the previous elements by self join.

Suppose that Lk-1 * Lk-1 is known as a Cartesian

product from Lk-1. This stage generates new candi-

date k-itemsets based on combining Lk-1 with them

which was found found in the previous iteration. Then

Ck denote candidate k-itemsets and Lk becomes the

frequent k-itemset.

2.2.3 Prune Test

Prunning eliminates several candidates from the

kitemset using the Apriori principle. The database

scanning process is carried out to determine the num-

ber of each candidate in Ck which will result in the de-

termination of Lk (that all candidates have an amount

less than the minimum amount of support). Repeat

steps 2 and 3 until no more sets of new candidates are

generated.

2.3 Previous Study

There have been many previous studies using Apri-

ori algorithms. There is study uses the itembased

collaborative ﬁltering method, where the system will

look for similarities in purchasing models with others.

Next, the system will search for ratings between items

based on the level of similarity that exists. After the

evaluation between pieces is obtaining, then this rat-

ing will be calculated similarity value between objects

using the Adjusted Cosine Similarity approach. The

results of the similarity calculations between items

will use for the next stage. This stage predicts a rating

that has never been done by a customer for a partic-

ular subject. This approach uses the Weighted Sum

formula whose prediction value will make a recom-

mendation to the customer (Kurniawan, 2016). The

application of a priori algorithm for movie website

recommendations is done by using a new approach

to adjust the features displayed and have an impact on

increasing the representative of the movie (Ma, 2016)

(Pal et al., 2017).

Apriori and Content-Based Filter (CBF) is also

used for determining the supply of compressor spare

parts (Kurniawati et al., ), market basket analysis in

the mini market (Elisa et al., 2018)(Mauliani et al.,

CONRIST 2019 - International Conferences on Information System and Technology

312

2015), sales analysis (Nursikuwagus and Hartono,

2016), positioning of products (Ningsih et al., 2016),

property search (Badriyah et al., 2018), game’s hard-

ware (Yanti et al., 2015), and product recommenda-

tion (Putra et al., 2019)(Kurniawan, 2016).

Also conducted research to design step-bystep

models to study the predictive analysis of Apriori al-

gorithms. From the weakness, namely speed, sev-

eral methods are proposed to overcome it, namely dy-

namic counting, determining the largest value for sup-

port and conﬁdence, and also taking large amounts of

dataset sample data (Ros¸ca and R

adoiu, 2015).

Apriori can also be combined with other meth-

ods. Combined Apriori with binary k-means at the

key parameter operating distance in the Central Air

Conditioning System (CACS). The results of this

study prove that priori discovered the rules of en-

ergy consumption. Operating data from equipment

was analyzed by DM, and cluster analysis was used

to improve the control effects of CACS (Yan et al.,

2019). Furthermore, there are Apriori combinations

and clustering algorithms that are used to get datasets

from trafﬁc accidents. With the help of a tool from

WEKA which provides many algorithms for data

mining selected a priori and clusters. From the test

results, it is proved that Apriori is better than the EM

cluster (Naﬁe Ali and Mohamed Hamed, 2018).

Content-Based Filter (CBF) and Content Filter-

ing (CF) combination were also conducted as hybrid

mechanism ﬁlters that can remove recommendations

that are not relevant to search keywords made by

mobile-based users. From the test results it is proven

that the proposed ﬁltering mechanism can improve

user personalization and enhance the ﬁltering expe-

rience on mobile devices (Zhao et al., 2015). Apri-

ori and Content-Based Filter (CBF) are not only used

for recommendations. There is research on improving

network sensors by modifying the priori algorithm.

From the test results it is known that modiﬁed algo-

rithms can better combine Wireless Sensor Network

(WSN) models from mobile nodes, also get greater

network coverage (Ji and Zhang, 2018). Then a study

was conducted to improves GPU performance with

the Apriori algorithm. From the test results it was

proven that GPU Apriori was able to improve the ef-

ﬁciency of item set mining (Jiang et al., 2017).

3 RESEARCH METHODOLOGY

This research consists of four stages, namely needs

analysis, algorithm analysis, implementation, and

testing.

3.1 Needs Analysis

At this stage, we do some data collection and needs

analysis related to the system being built is carried

out. The data is collected from books and articles

from journals or proceedings. Needs analysis in-

cludes the speciﬁcations of the system to be built and

the computer used.

3.2 Algorithm Analysis

At this stage we learn and ﬁnd the ways of working,

how to calculations, and learn the ﬂow process of the

Apriori algorithm that implemented in the mobile ap-

pllication.

3.3 Implementation

The results of the learning of the Apriori algorithms

are then applied in the Android system. To help to un-

derstanding, case studies were given on the android-

base shoe ad application.

3.4 Testing

After the algorithm is applied, then the application

is tested to determine the performance of the Apri-

ori algorithms that implemented and performance of

the application. The tests carried out are algorithm

performance testing using whitebox and blackbox

method.

4 RESULTS AND DISCUSSIONS

4.1 Problems Analysis

The types and brands of shoes in shoe stores continue

to increase each year to meet the different interests

of customers. But with the varying number of shoe

products, customers need a long time to choose the

desired shoes.

4.2 Proposed System Analysis

The explanation of the proposed system analysis will

be explained using use case diagrams, activity dia-

grams, and class diagrams, which can see below:

4.2.1 Use Case Diagram

Based on Figure 1, it can explain that in the system

there are two actors, namely the user and admin.

Product Recommendation Systems using Apriori in the Selection of Shoe based on Android

313

space

Figure 1: Use case diagram

4.2.2 Activity Diagram

An activity diagram is a diagram that describes the

ﬂow of functionality from the system. The types

of activity diagrams this time are registers, logins,

change passwords, log out, add advertisements, delete

ads, add categories, delete categories, and activity di-

agrams select categories. For example, can be seen in

Figure 2, for activity logins diagrams.

Figure 2: Login processes Activity Diagram

4.2.3 Class Diagram

Class diagrams are diagrams to display several classes

in the software system that will develop. The pro-

posed class diagram can see in Figure 3.

space

Figure 3: Class diagram

4.3 Discussion of Apriori Algorithm

Apriori algorithms are using so that computers can

learn the rules of the association, look for patterns

of relationships between items in data so that they

can make a product recommendation. An example of

transaction data can see in Figure 4.

Figure 4: Transaction data.

The steps of using a priori algorithm can see in the

explanation below:

4.3.1 First Iteration

In ﬁrst iteration, calculate the number of transactions

for each item. Where the method of calculating sup-

port can see in equation (1).

Support(A) =

Numbero ftransactionscontaining(A)

TotalTransactions

x100 (1)

The results of calculations in ﬁrst iteration can be

seen in Figure 5.

Figure 5: First iteration results.

CONRIST 2019 - International Conferences on Information System and Technology

314

4.3.2 Second Iteration

In second iteration, do a combination of the previous

k-itemset, where the method of calculating support

can see in equation (2).

Support(A, B) = P(A ∩ B)

Numbero f transactioncontaining(AandB)

TotalTransactions

x100

(2)

The results of calculations in the second iteration

can see in Figure 6. Support Count is obtained from

shoe transaction data that have been collected safely

in 2018, namely 24 transactions. Then it is sought a

transaction involving two brands of shoes. In the elab-

oration of Figure 6, the ﬁrst point mentioned that Adi-

das and Nike have a support count of 8 transactions.

This means that out of the 24. transactions there are 8

transactions that intersect between Adidas and Nike.

Figure 6: Second iteration results.

4.3.3 Establishment of Associative Rules

Calculate conﬁdence using a formula that can see in

equation 3.

Con f idence = P(B|A) =

Numbero f transactioncontaining(AandB)

Totalo f transactioncontaining(A)

x100 (3)

The results are shows in Figure 7.

Figure 7: Associative rules results.

4.4 Implementation

Screen design is the ﬁrst display that is in the applica-

tion for the admin. The following Figure 8 is a screen

design minimum support for database sales.

Figure 8: Minimum Support for database sales

Another screen design, can see in Figure 9, it is

about item formation results from the database. Fig-

ure 10 shows a display of determining minimum pa-

rameters and Figure 11 is about the results of the as-

sociation carried out from Apriori algorithm.

Product Recommendation Systems using Apriori in the Selection of Shoe based on Android

315

space

Figure 9: Item formation results from the database

Figure 10: display of determining minimum parameters

space

Figure 11: display of the results of the association carried

out a priori based.

4.5 Testing

To ﬁnd out the performance of the application and the

algorithm that has been made, then testing is done us-

ing Whitebox and Blackbox method. The whitebox

used to testing based on manual calculation and the

application calculation as shows in Figure 12, and the

Blackbox testing used to measure the perform of the

application and shows in Figure 13.

Both testing methods are considered to have rep-

resented the purpose and utilization of the recommen-

dations made.

CONRIST 2019 - International Conferences on Information System and Technology

316

space

Figure 12: Whitebox Testing Manual Calculation.

Figure 13: Blackbox Testing.

5 CONCLUSIONS

The results of the study are looking for the level of

support and conﬁdence in shoes so that shoes are

obtained that are more accurate using data mining.

Based on testing the system uses the apriori method

to produce rules derived from the pattern of the com-

bination of two items. From the calculation results

obtained that the highest associative rule product is

New Balance → Adidas with a value of 87.5% and

the lowest result on the Adidas → Puma with a value

of 18.2%. The rules above consist of Antecedent is a

form of a condition rather than rules, a consequent is

a form of a statement rather than rules, lift shows the

level of power rules random events of antecedent and

consequent based on their respective support.

Further research can carry out a variety of addi-

tional testing to test the performance of the apriori al-

gorithm that created and implemented. Besides that, it

can also combine or compare apriori algorithms with

other recommendation algorithms.

REFERENCES

Aceto, G., Ciuonzo, D., Montieri, A., and Pescap

e, A.

(2018). Multi-classiﬁcation approaches for classify-

ing mobile app trafﬁc. Journal of Network and Com-

puter Applications, 103:131–145.

Aceto, G., Ciuonzo, D., Montieri, A., and Pescap

e, A.

(2019). Mobile encrypted trafﬁc classiﬁcation us-

ing deep learning: Experimental evaluation, lessons

learned, and challenges. IEEE Transactions on Net-

work and Service Management, 16(2):445–458.

Badriyah, T., Azvy, S., Yuwono, W., and Syarif, I. (2018).

Recommendation system for property search using

content based ﬁltering method. In 2018 Interna-

tional Conference on Information and Communica-

tions Technology (ICOIACT), pages 25–29. IEEE.

Elisa, E. et al. (2018). Market basket analysis pada

mini market ayu dengan algoritma apriori. Jurnal

RESTI (Rekayasa Sistem dan Teknologi Informasi),

2(2):472–478.

Febriansyah, R. and Samsinar, S. (2018). Desain sis-

tem informasi pembelian dan penjualan dengan uni-

ﬁed modeling language (uml) tools pada toko simple-

bag. IDEALIS: InDonEsiA journaL Information Sys-

tem, 1(1):7–11.

Ji, Q. and Zhang, S. (2018). Research on sensor net-

work optimization based on improved apriori algo-

rithm. EURASIP Journal on Wireless Communica-

tions and Networking, 2018(1):1–9.

Jiang, H., Xu, C.-W., Liu, Z.-Y., and Yu, L.-Y. (2017). Gpu-

accelerated apriori algorithm. In ITM Web of Confer-

ences, volume 12, page 03046. EDP Sciences.

Kasemin, H. K. et al. (2016). Agresi Perkembangan

Teknologi Informasi. Prenada Media.

Kavitha, M. and Selvi, S. T. (2016). Comparative study on

apriori algorithm and fp growth algorithm with pros

and cons. International Journal of Computer Science

Trends and Technology (I JCS T)–Volume, 4.

Kurniawan, A. (2016). Sistem rekomendasi produk sepatu

dengan menggunakan metode collaborative ﬁltering.

In Seminar Nasional Teknologi Informasi dan Komu-

nikasi, pages 610–614.

Kurniawati, L., Kusuma, A. E., and Dewansyah, B. Im-

plementasi algoritma apriori untuk menentukan perse-

diaan spare part compressor. Computer Engineering,

Science and System Journal, 4(1):6–10.

Product Recommendation Systems using Apriori in the Selection of Shoe based on Android

317

Ma, K. (2016). Content-based recommender system for

movie website.

Mauliani, A., Hartati, S., and Musdholifah, A. (2015). Pem-

bentukan temporal association rules menggunakan al-

goritma apriori (studi kasus: Toko batik diyan solo).

IJCCS (Indonesian Journal of Computing and Cyber-

netics Systems), 10(1):71–80.

Naﬁe Ali, F. M. and Mohamed Hamed, A. A. (2018). Usage

apriori and clustering algorithms in weka tools to min-

ing dataset of trafﬁc accidents. Journal of Information

and Telecommunication, 2(3):231–245.

Ningsih, W. et al. (2016). Implementasi data mining un-

tuk penentuan posisi barang pada rak menggunakan

metode apriori pada pt midi utama indonesia. Techno.

Com, 15(4):335–342.

Nursikuwagus, A. and Hartono, T. (2016). Implemen-

tasi algoritma apriori untuk analisis penjualan dengan

berbasis web. Simetris: Jurnal Teknik Mesin, Elektro

dan Ilmu Komputer, 7(2):701–706.

Pal, A., Parhi, P., and Aggarwal, M. (2017). An im-

proved content based collaborative ﬁltering algorithm

for movie recommendations. In 2017 tenth interna-

tional conference on contemporary computing (IC3),

pages 1–3. IEEE.

Putra, A. A. C., Haryanto, H., and Dolphina, E. (2019).

Implementasi metode association rule mining dengan

algoritma apriori untuk rekomendasi promo barang.

CSRID (Computer Science Research and Its Develop-

ment Journal), 10(2):93–103.

Raharjana, I. K. (2017). Pengembangan Sistem Informasi

Menggunakan Metodologi Agile. Deepublish.

Rosca, D. G. and Radoiu, D. (2015). Step-by-step model for

the study of the apriori algorithm for predictive analy-

sis. Scientiﬁc Bulletin of the Petru Maior” University

of Targu Mures, 12:44–47.

Yan, L., Qian, F., and Li, W. (2019). Research on key pa-

rameters operation range of central air conditioning

based on binary k-means and apriori algorithm. Ener-

gies, 12(1):102.

Yanti, N., Rahmi, R., and Ruliah, R. (2015). Penerapan

algoritma collaborative ﬁltering untuk rekomendasi

games hardware. Jutisi: Jurnal Ilmiah Teknik Infor-

matika dan Sistem Informasi, 2(1).

Zhao, F., Yan, F., Jin, H., Yang, L. T., and Yu, C. (2015).

Personalized mobile searching approach based on

combining content-based ﬁltering and collaborative

ﬁltering. IEEE Systems Journal, 11(1):324–332.

CONRIST 2019 - International Conferences on Information System and Technology

318