Comprehensive Approach to Assure the Quality of MSME Data in

Indonesia: A Framework Proposal

Mujiono Sadikin

, Adi Trisnojuwono

, Rudiansyah

and Toni Widodo

Faculty of Computer Science, Universitas Bhayangkara Jakarta Raya, Kota Bekasi, Indonesia

Deputy of Enterpreneurship, Ministry of Coperation and Small Medium Enterprise, Jakarta, Indonesia

Mitreka Solusi Indonesia, Jakarta, Indonesia

Keywords:

MSME, Data Quality, Data Cleansing, Attribute Domain Constraint, Relation Integrity Rule.

Abstract:

As mandated by the laws and regulations that have been released, the Government of Indonesia has decided

that Cooperation and MSME empowerment policies must be determined based on accurate Cooperation and

MSME data proﬁles. Therefore, the Government of Indonesia, in this case the Ministry of Cooperation SMEs,

executes a complete data collection program of Cooperation and MSME proﬁle. Due to the characteristics and

constraints of data collection, many risks must be mitigated. The main risk identiﬁed in this program is the

possibility of reduced data quality for Cooperation and MSME caused by some factors. This paper presents

a proposed comprehensive framework to ensure the quality of Cooperation and MSME data based on Khan’s

data quality criteria previously deﬁned. The aim of the proposed framework is to prevent, detect, repair, and

recover dirty data to achieve the required minimum standards of data quality. The proposed framework covers

all stages and aspects of the data collection process. In the data cleaning and correction stage, we investigate

on many techniques namely rule based, selection based, and machine learning based. In the initial validation

of the framework presented in this paper, the results of several data cleansing methods applied are discussed.

1 INTRODUCTION

Due to the importance of data quality in all human

live ﬁeld, the management of data quality studies

has got more attention from many researchers and

practitioners as well (Liu et al., 2018). Data qual-

ity is a crucial issue must be tackled. Many re-

searchers have published the results of their studies

to address data quality issues speciﬁcally in certain

areas. Several publications on data quality include:

in the health sector (Dziadkowiec et al., 2016), on-

line sales and marketing sector (Arndt et al., 2022),

company registration cases in several countries (Niki-

forova, 2020), environment monitoring (Zhang and

Thorburn, 2022), concrete analysis in infrastructure

sector (Ouyang et al., 2021), and manufacture indus-

try (Martinez-Luengo et al., 2019).

Many approaches to assure the data quality ful-

ﬁll the requested standard have been proposed. Lie et

al summarised those approaches as published in (Liu

et al., 2018). In their studies authors also investigate

the application of some techniques and algorithms in

data cleansing that are performed in various analy-

sis stages according to data type. Based on the three

previous general methods, the authors make improve-

ments by adding visualization block. Despite the ad-

vantages of the proposed framework, the framework

has not been validated to the real data case. A data

quality improvement technique that combines Kahn’s

work and SPSS query syntax was proposed by Dzi-

adkowiec et al (Dziadkowiec et al., 2016). By uti-

lizing the Kahn Data Quality Framework, the authors

claim that the quality of the resulting data can be bet-

ter. In this paper, we adapt Kahn framework as part

of the MSME data quality assurance framework by

elaborating the possibility of implementing ML and

DL approaches.

Another approach regarding quality data improve-

ment is proposed by (Arndt et al., 2022). In their

study, authors use ﬁlters to improve the quality of on-

line market analysis data. The data used in this exper-

iment was provided through a crowdsourcing service,

i.e. Mturk. The aim of this methodology is to com-

pare each of the four ﬁlter categories based on the data

sources. Those four used ﬁlterers are direct selection,

direct accuracy, statistical selection, and statistical ac-

curacy.

User oriented data-driven methods is proposed by

154

Sadikin, M., Trisnojuwono, A., Rudiansyah, . and Widodo, T.

Comprehensive Approach to Assure the Quality of MSME Data in Indonesia: A Framework Proposal.

DOI: 10.5220/0012445700003848

Paper published under CC license (CC BY-NC-ND 4.0)

In Proceedings of the 3rd International Conference on Advanced Information Scientiﬁc Development (ICAISD 2023), pages 154-161

ISBN: 978-989-758-678-1

(Nikiforova, 2020). This approach contains three

components, namely data objects, data quality speciﬁ-

cations, and data quality measurement processes. The

proposed model is applied to open data set of register

companies in several countries such as Latvia, Nor-

way, England, and Estonia. The authors claim that the

proposed method is able to overcome the weaknesses

left by the previous approach. Despite the proposed

method is quite comprehensive, it only handles data

that meets certain criteria such as: complete, free of

ambiguous values, and correct. This such constraints

present that the data quality issues continuously pro-

vide new challenges must be addressed.

Another aspect that must be considered to ensure

data quality is the proper handling of master data

(Prokhorov and Kolesnik, 2018). The proposed mas-

ter data model management system consists of three

activities: consolidation, harmonization, and man-

agement. In the consolidation stage, improvements

are applied to the data structure and data collection.

In the harmonization stage, alignment, normalization

and classiﬁcation are carried out, whereas in the man-

agement action stage that must be carried out is cen-

tralization and management.

Data quality assurance is more challenging when

data collection processes are almost real time, such

as highfrequency water quality monitoring systems

as investigated by Zhang et al (Zhang and Thorburn,

2022). Many factors inﬂuence the decreasing of the

real time data quality such as network problems, de-

vice malfunction, device replacement etc. To over-

come the missing values of real time data set that

will affect the quality of the information provided,

the authors developed a cloud-based system that com-

bines several techniques and advance algorithms to

perform the missing values imputation. Several im-

putation techniques used in the system such as Mean

Imputation, LOCF, Linear Imputation, EM, MICE,

Dual-SSIM and M-RNN. Among these techniques,

by overall Dual-SSIM provides the best performance

when it applied to nitrate and water temperature data.

Despite the methods is powerful in handling the real

time data, however it does not handle integrity be-

tween data or integrity between attributes / attributes

relations.

Real world data is mostly dirty due to errors found

in data sets. In many cases, the actual data set is

inconsistent, contains missing values, lacks of in-

tegrity, ambiguous, and contains outliers. Therefore

data cleaning is not only the main task, but also the

most important part of data management since data

quality determines the quality of the information pro-

duced (Ridzuan and Zainon, 2019). The handling

of data cleansing cases requires a different approach.

Ouyang et al, as presented in (Ouyang et al., 2021),

used a ML-based ensemble approach to detect outliers

in a concrete measurement regression data set. The

technique used in this case is the ANN-based model

compared to KNN, LOF, COF, OCSVM, IFOREST,

ABOD, and SOS. The approach used in the study is

the best algorithm selection approach with forward

and backward techniques. Based on the experimental

results, ANN gives the best results in detecting out-

liers in the regression data used.

Noisy data that contributes to unreasonable

decision-making also occurs in the energy manufac-

turing industry such as the offshore wind turbine

structures health data collected through the SCADA

monitoring system. To overcome the problem,

(Martinez-Luengo et al., 2019) proposes a method

based on ANN techniques to improve data quality

through automation of data cleaning processes. The

proposed framework consists of two steps: data noise

checking and removal, and missing data imputation.

This research was conducted to improve the quality

of fatigue assessment on turbines which are thought

to be heavily inﬂuenced by the quality of monitor-

ing data generated by SCADA sensors. Therefore, in

his research the authors compared the quality of data

without cleaning with the quality of data with clean-

ing using the proposed method. From the experimen-

tal results, the authors concluded that the quality of

the data after cleaning proved to be better.

In their review of big data cleansing, (Ridzuan and

Zainon, 2019) summarize some methods can be ap-

plied to the purposes. Those methods are developed

based on various techniques such as rule-based, ML-

based, and knowledge-based. However, those existing

methods contain some limitations when it deals with

dirty data.

The complete data collection of Cooperations and

SMEs conducted by the Ministry of SMEs is a unique

data collection model. The uniqueness, complexity,

and problems are found in all components, including

area coverage, individual data targets, data collection

model which is performed manually, various skill and

knowledge of data collection ofﬁcers, the complex-

ity of data entry forms, short time allocated, and the

project management as well. In terms of area cover-

age, the project covers more 240 district, 34 provinces

crossover Indonesia Country with various topogra-

phies and land contours. The data collection is car-

ried out manually by more than 1.000 enumerators.

As other models of real data collection, data qual-

ity is also a major issue that must be resolved before

further use of the data. Due to this uniqueness and

exclusiveness, to the author’s knowledge, there is no

model/approach that can deal with this data quality

Comprehensive Approach to Assure the Quality of MSME Data in Indonesia: A Framework Proposal

155

standard problem comprehensively. Therefore, this

paper presents a proposed framework for cleaning the

data obtained by this kind of collection model. In

general, the proposed framework consists of compo-

nents/steps speciﬁc to the case and a combination of

existing/published approaches. The proposed frame-

work also consists of various approaches according to

the case found.

Figure 1: The distribution of City/Regency of data collec-

tion area.

2 RESEARCH METHOD

2.1 Overview of MSME Data Collection

Program

This paper presents a framework proposal and case

study of data cleansing in the Complete Data Collec-

tion performed by Ministry of Cooperation and MSE.

This section discuss the overview of the program i.e.

data collection object, its attributes, and the issues

have to be resolved.

The objects of the data collection are the business

actors of cooperation , micro, small and medium en-

terprises. In accordance with the mandate of the laws

and regulations, the provision of single data for this

business group is the duty and authority of the Min-

istry of Cooperations and SMEs (Indonesia, 2021).

To carry out the duties and responsibilities of devel-

oping a single data for Cooperations and MSMEs, in

the year of 2022 data collection is performed for more

than 9 million individual Cooperation and MSMEs

data by name by address. In terms of business lo-

cation, micro and small business actors are very dy-

namic and vary. In the year stage, the data collection

is limited to business actors who occupy a permanent

business location.

For data collection purposes, each individual data

is identiﬁed by 237 attributes which are divided into

15 blocks (groups) of attributes. Those attribute data

blocks are: identity of place of business, identity

of business characteristics, identity of business ac-

tors, business licenses, awards received, raw mate-

rials, production, labor, production processes, part-

nerships, ﬁnancial business, coaching ever received,

additional notes , and information from the registrar.

The number of attributes in each block varies from 10

to 53.

The data collection area covers 34 Provinces and

240 Regencies/Cities/Districts in Indonesia. The dis-

tribution of districts/regencies/cities for data collec-

tion in all Indonesia Provinces is presented in Fig. 1.

For each province, based on considerations of ease

of transportation access and adequacy of infrastruc-

ture, there were chosen between 1 (one) to 33 re-

gencies/districts/cities of each province. The charac-

teristics of regencies/cities in the Indonesian territory

widely vary in terms of topography, availability of in-

frastructure, social economy, and level of education

of business actors. In the ﬁrst stage of this data col-

lection, districts/cities were selected based on consid-

erations of the populationof business actors, afford-

ability, and availability of data communication infras-

tructure.

Data collection was carried out by enumerators.

The number of enumerators varies from one dis-

trict/city to another, from 1 to. 1.709 enumerators.

The parameters for the number of enumerators are

based on consideration of the estimated number of

business actors in the district/city concerned, the size

of the coverage area, and the degree of access difﬁ-

culty. This data collection also requires coordination

across ministries/agencies both at the central and re-

gional levels considering that the development of Co-

operation and MSME business actors is cross-sector.

Complete data collection is done by utilizing web-

based applications and mobile apps that are accessed

using the internet. This application is speciﬁcally pre-

pared for -one of them - the implementation of this

data collection.

Given the complexity of the complete Coopera-

tion and MSME data collection program both in terms

of enumerators, data collection areas, and the num-

ber of parties involved, the potential for data col-

lection errors is quite large. Enumerators (data col-

lectors) come from various educational and cultural

backgrounds. Even though training and socialization

have been carried out regarding the process and how

to use the application for complete data collection,

due to different levels of expertise, the potential for er-

ror entry by enumerators still exists. Each district/city

is given a quota/target for a certain amount of indi-

vidual data which is then distributed to enumerators.

The minimum target that must be achieved also has

the potential to cause enumerators to make mistakes

in data collection, either intentionally or unintention-

ally. Regencies/cities for data collection also have

different characteristics, both in terms of geographic

topology and available information technology facili-

ICAISD 2023 - International Conference on Advanced Information Scientiﬁc Development

156

ties, in this case internet connection. An unstable in-

ternet connection allows for errors in data collection,

while transportation difﬁculties impact the enumera-

tor’s home range.

Considering that Cooperation and MSME data

meet minimum quality, both in terms of ﬁlling in val-

ues on attributes, as well as the amount of data while

the potential for data errors is quite large, a proper

mechanism is needed to ensure data quality is fulﬁlled

the requirements. In order to assure the quality of the

data, in this case study we propose a data quality as-

surance framework based on the characteristics of the

data collection project and data collected as well. This

framework can also be an alternative for data collec-

tion which is massive in nature, involves many par-

ties, within a narrow time span, involves many parties

and covers a wide and varied area.

2.2 Proposed Framework

2.2.1 Overview Cooperation and MSME Data

Attributes

The purpose of this data collection is to produce qual-

ity and sufﬁcient data both in terms of individual data

amount and attributes value that will be used as mate-

rial for formulating policies on empowering Cooper-

ation and MSMEs. For this purpose, according to the

mandate of PP No. 7 of 2021 Article 55 Paragraph

(3) (Indonesia, 2021) it is required that the main data

variable group in the Single Data Information System

(Sistem Informasi Data Tunggal/SIDT) of Coopera-

tion and MSMEs contains at least Business Identity

and Business Actors. In its elaboration, the Min-

istry of Cooperation and SMEs breaks down these

variables into attributes grouped into Business Ac-

tor Identity, Business/Business Entity Identity, Gen-

eral Business Characteristics, Human Resources, Pro-

duction/Business Processes, Marketing, and Financial

Status. Other attributes as additions include: prod-

uct/service marketing area, type of workforce, suppli-

ers, turnover, venture capital, and other attributes.

2.2.2 Quality Data Assurance Framework

In compiling the framework, we use two references,

i.e. data requirements speciﬁcation references and

data quality regulatory requirements references. The

reference of data requirements is Government Regu-

lation no. 7 of 2021, whereas the reference for the

quality of data fulﬁllment is based on ﬁve data quality

rules according to Kahn (Kahn et al., 2012). The as-

surance data quality framework for Cooperation and

SMEs is depicted as Fig. 2. The objective of complete

data collection on Cooperation and SMEs is the ful-

ﬁllment of the Data Quality Rule which consists of 5

rules. Data quality assurance efforts are implemented

at each stage of data collection starting from prepa-

ration, implementation/execution, to ﬁnal data collec-

tion. The data quality assurance function includes all

necessary actions, namely prevention of possible data

errors/abnormalities, anticipation of possible errors,

detection of data errors and correction of erroneous

data.

In each action the mechanism or tools used are

in accordance with the context of the action. In the

preparation stage, features/functions are implemented

in the application that are used to maintain the quality

of ﬁeld values since they were originally entered. To

minimize data errors due to human error, in addition

to implementing application features/functions, dur-

ing the preparation stage, intensive training was also

carried out for enumerators and veriﬁers.

At the implementation stage, anticipatory ac-

tions of errors are carried out by implementing two

stages of veriﬁcation. The ﬂow and the second

phase of the veriﬁcation process are carried out semi-

automatically using an application. The ﬁrst stage of

veriﬁcation is carried out by the enumerator coordi-

nator, while the second stage of veriﬁcation is carried

out by veriﬁcation ofﬁcers at the regency/city level.

Figure 2: Proposed framework of Cooperation & MSME

data quality assurance.

After the data has been collected and veri-

ﬁed at these two stages, actions are taken to de-

tect/identify inappropriate or abnormal data and cor-

rect errors/abnormalities. The mechanisms used at

this stage are: assistance with unusual data using sta-

tistical techniques, SQL-based queries, and veriﬁca-

tion by experts in the ﬁeld of microeconomics, espe-

cially Cooperation and SMEs. Detection of data that

may not be normal is also carried out using ML tech-

niques, in this case is clustering.

Data correction measures to ensure that data is

correct and accurate are carried out using rule-based

techniques and approaches based on ML and DL. The

potential use of ML/DL is for example to assess the

suitability of a place of business with the narrative of

a business sector or class of business. The process

of collecting and assuring data quality is presented in

Figure 3.

Comprehensive Approach to Assure the Quality of MSME Data in Indonesia: A Framework Proposal

157

Figure 3: Flow of data quality assurance process.

3 RESULTS AND DISCUSSION

Currently, all stages of data collection implementa-

tion for this year’s program have been completed. The

total instant data collected is more than 9,000,0000.

Assurance of data quality as in the proposed frame-

work is already at the Data Cleansing stage, namely

detective and correction. Some of the detection and

correction processes have already been implemented,

while other ML/DL/AI based processes are in the de-

sign stage. This section presents the results of apply-

ing some of the mechanisms and tools to the actions

of the proposed framework.

3.1 Preparation

3.1.1 Application Function/ Feature

Attribute domain constraint To ensure compliance

with ADC rules, we apply mandatory ﬁeld features

and value constraints for attributes which are the

minimum requirements according to laws and reg-

ulations. Of the 237 attributes, 42 attributes are

mandatory that must be ﬁlled. If these attributes

are not ﬁlled, data collection cannot be continued.

The mandatory attribute groups includes the busi-

ness actor identity attribute, company characteristics,

business location identiﬁcation, business production,

workforce, and production process.

The mandatory business actor identity attribute

groups include the attributes of the entrepreneur’s

name, gender, disability (y/n), entrepreneur’s sta-

tus, Entrepreneur’s NIK (citizen identity number),

whether the address of the place of business is the

same as the entrepreneur’s address, province, dis-

trict/city, sub-district, kelurahan/village/ nagari, com-

plete address, telephone/cell phone number, Whats

App, entrepreneur education, are you a member of a

Cooperation, what type of Cooperation do you join,

do you have other jobs, what other types of work

do you have. The mandatory business characteris-

tics group includes the main business/company activ-

ity attributes, the main products (goods or services)

produced/sold, business entity status, initial capital at

establishment, and date of operation. Mandatory busi-

ness place attribute groups are province, district/city,

sub-district/district, sub-district/village/nagari, full

name of business/company, name of commer-

cial/popular business, place of business, and busi-

ness address. Mandatory business production at-

tribute groups are production of goods/services pro-

duced, marketing of goods/services produced, mar-

keting methods used, and address. The mandatory

attributes of the workforce group are: wages and

salaries, other incentives, number of months worked

in one year, average working days per month, and av-

erage hours worked per day. Meanwhile, the produc-

tion process group only attributes the use of technol-

ogy in the production process.

At this stage, the guarantee of the fulﬁllment of

the ADC is also carried out by checking by the sys-

tem for attribute values that must meet certain criteria.

Examples of such this attributes and their limitations

are presented in Table 1.

Table 1: Attributes and their domain constraint.

Attribute Domain Constraint

NIK (Citizen ID) Numeric, 14 digit

No HP Numeric, 9 – 11 digit

Kode Pos (Zip Code) Numeric, 5 digit

Nama Pelaku Usaha

(Business Actor

Name)

Alphabet

Kategori Usaha

(Business Category)

One Alphabet, refer

to KLUI (Indonesia

Business Field Cat-

egory) (Keuangan,

2012)

Attribute Dependency Integrity Rule The follow-

ing Table 2 presents a list of attribute dependency in-

tegrity guarantees implemented in data collection ap-

plications. The right side is an attribute that must meet

the dependency rule on the left attribute. For example,

the province must match the proﬁle of the enumerator

in charge of the province concerned.

3.1.2 Execution

At the data collection/data collection execution stage,

the anticipation that is carried out to minimize data er-

ICAISD 2023 - International Conference on Advanced Information Scientiﬁc Development

158

Table 2: Attribute dependency integrity rule.

Attribute Dependency Rule

Enumerator Proﬁle –>(Province,

/City/District )

Zip Code –>Sub District

NIK (citizen identity number) –>Gender

Employee Number –>Business Status

Cooperation Member –>The type of Co-

operation involved

Other business –>Profession

Finance statement –>This year income

Marketing utilizes digital media–>type of

media digital used

If the business category is ”G” –>( Sales

income, Purchase prices of goods sold,

Commission on net consignment sales)

rors/inaccuracies is to apply collection payment rules

and veriﬁcation. Payments to enumerators for their

data collection services are based on the units of ver-

iﬁed data they can collect. Two step of veriﬁcation

is implemented in stages. The ﬁrst step of veriﬁcation

was carried out by the enumerator coordinator and the

second-level veriﬁcation is carried out by employees

of the relevant ofﬁces in City/District Governments,

as presented in Figure 2.

3.2 Collection

3.2.1 Missing Value Detection

Missing value detection is performed by using SQL

query tools. As the results of database query exe-

cution, it was found approximately 6,850,000 miss-

ing value. The three attributes with the highest miss-

ing values were found in 2,519,084 of marketing

method attribute, 1,606,018 of telephone number, and

1,455,625 employee. Other attributes that dominate

the missing value are the business production cate-

gory, business location, and business capital.

3.2.2 Expert Veriﬁcation of Attribute Domain

Constraint

The experts who carry out the veriﬁcation at this stage

are former employees of the Central Bureau of Statis-

tics who are proﬁcient in the concept of microeco-

nomics. At this veriﬁcation stage, it is carried out

using statistical tools to identify the mean, median,

minimum value and maximum value of several impor-

tant attributes. Despite it has been implemented the

two levels of veriﬁcation in the execution phase, there

are still many attributes with unrealistic values. Many

cases of unrealistic values occur in business aspects

such as working capital, annual business turnover, and

expenses. For such those unrealistic data, the expert

establish the veriﬁcation rule based on the expertise,

regulation, and experiences as well. Some of the ver-

iﬁcation results is presented as Table 3.

Table 3: Expert veriﬁcation result of attribute domain con-

straint.

Block

Question

Attribute

Category

Business Category Normal Value

Block 8 Number of

Employee

Individual Business Minimum = 1

Maximum = 20

Non Individual (Com-

pany) Business

Minimum = 20

Maximum = 100

Employee

Salary

(Rp.)

Individual Business Minimum =

700.000

Maximum =

3.100.000

Non Individual (Com-

pany) Business

Minimum =

2.300.000

Maximum =

6.300.000

3.2.3 Expert Veriﬁcation of Integrity

Constraints

Based on the results of the veriﬁcation of experts, it

was found that there was a discrepancy in the value of

integrity between attributes that should be consistent.

The ﬁndings of this integrity inconsistency occur in

attributes related to ﬁnance and business. The main

ﬁndings for this case are presented as a Table 4.

3.3 Data Correction

In the stages we have performed data correction based

on rule deﬁned and expert justiﬁcation. All of data

error, inconsistent, or missing value as described in

sub section Missing Value Detection and Expert Ver-

iﬁcation of Attribute Domain Constraint have been

corrected. Currently we are preparing to utilize the

ML/DL approach to assist the error detection and cor-

rections. Table 5 presents the initial identiﬁcation of

ML/DL approaches to be applied on the objective of

qualiﬁed data rules.

Comprehensive Approach to Assure the Quality of MSME Data in Indonesia: A Framework Proposal

159

Table 4: Attribute Dependency Integrity Rule.

Integrity Issues # of Instant

Data

Business Category is indi-

vidual but its business capi-

tal >Rp. 1 B

2.725

Business Category is corpo-

rate but its business capital

<500 million

6.818

Business Category is Cor-

porate but its yearly omzet

<25.000.000

3.220.942

Business Category is indi-

vidual but its each employee

monthly salary <Rp.

700.000

8.52.145

Table 5: ML/DL Approach Identiﬁcation.

Attribute Objec-

tive

L/DL Task

(Target)

Tools Predictor

Business

Cate-

gory

ADC,

RIC

Classiﬁcation,

new- line

Text Gen-

erator (Text

indicates

Business

Category)

NN-

RNN

GAN-

RNN

Business Image Photo,

Address

Business

Type

RIC Classiﬁcation

(Indi-

vidual,

Corporate)

Classiﬁer

(De-

cision

Tree,

RNN-

CNN)

Business Image Photo,

Financial Aspects at-

tributes

Working

Capi-

tal

RIC Classiﬁcation

(Range of

WC)

Classiﬁer

(Multi

Class,

Deci-

sion

Tree,

SVM,

NB)

Omzet, Number Of

Employees, Busi-

ness Category, Ad-

dress/Location, Prod-

uct, Market Segment,

Target Market, product

main raw materials

Type

Target

Mar-

ket

RIC Classiﬁcation

(Type of

TM)

Classiﬁer

(Multi

Class,

Deci-

sion

Tree,

SVM,

NB)

Market Segment, Tar-

get Market, product

main raw materials,

working capial, market-

ing methods, Busi- ness

Category, Employee

(num- bers, education,

salary,... )

Remark

The objective is to generate caption describes busi-

ness category based on the activity/location business

photo image. Many studies presents these approach

are powerful to this task (Aghav, 2020; Ghandi et al.,

2022; Srivastava et al., 2022) CNN based clas- siﬁer

is the most popular technique for object-image recog-

nition (Xuefeng Jiang, 2019; Luo et al., 2020; Lee

et al., 2018; Srivastava et al., 2022; Sadikin et al.,

2020; Elngar et al., 2021), where as RNN is suitable

applied to a sequence data set (Olah, 2015; Sastrawan

et al., 2022; Sadikin et al., 2016) For the classiﬁcation

task due to its predictors is quite simple, the conven-

tional classiﬁers are powerful enough to overcome

the “business-like” problem as pre- sented in (Sadikin

et al., 2021).

4 CONCLUSIONS

The aim of the proposed framework presented in the

article is to assure the quality data of Cooperation and

MSMEs as the result of Complete Data Collection

program run by Ministry of Cooperation and SME

full the minimum standard required. The proposed

framework covers all stage in the Data Collection

Project by performing all activities required i.e antic-

ipation, prevention, detection, and correction. Cur-

rently some of the stages and activities contained in

the framework have been performed. As the results of

the application of some parts of the framework, more

than 6.8 million error data are detected and corrected.

As the Complete Data Collection Program which

is continuously carried out to achieve 65 million Co-

operation and MSME individual data, the assurance

of data quality standard will be more complex. There-

for, we elaborate the utilization of AI approaches i.e.

ML / DL. Some problems of data quality and its

ML/DL approach to address the obstacle are also pre-

sented in the paper. In the next short time, we will

investigate the AI approach to be applied in the pro-

gram.

ACKNOWLEDGEMENTS

Thank the Deputy of Entrepreneurship Ministry of

Cooperation and MSE who give us the permit the pub-

lish the lessons learned and the data cleansing case of

the Program.

REFERENCES

Aghav, J. (2020). Image captioning using deep learning.

International Journal for Research in Applied Science

and Engineering Technology, 8:1430–1435.

Arndt, A., Ford, J., Babin, B., and Luong, V. (2022). Col-

lecting samples from online services: How to use

screeners to improve data quality. International Jour-

nal of Research in Marketing, 39:117–133.

Dziadkowiec, O., Callahan, T., Ozkaynak, M., Reeder, B.,

and Welton, J. (2016). Using a data quality frame-

work to clean data extracted from the electronic health

record: A case study. In eGEMs (Generating Evidence

Methods to improve patient outcomes) 4, volume 11.

Elngar, A., Arafa, M., Fathy, A., Moustafa, B., Mahmoud,

O., Shaban, M., and Fawzy, N. (2021). Image classi-

ﬁcation based on cnn: A survey. Journal of Cyberse-

curity and Information Management, 6:18–50.

Ghandi, T., Pourreza, H., and Mahyar, H. (2022). Deep

learning approaches on image captioning: A review.

Tech. rep. arXiv:2201.12944.

ICAISD 2023 - International Conference on Advanced Information Scientiﬁc Development

160

Indonesia, G. (2021). Indonesia goverment law number

07 year 2021 of ease, protection, and empowerment

of cooperations and micro, small, and medium enter-

prises.

Kahn, M., Raebel, M., Glanz, J., Riedlinger, K., and Steiner,

J. (2012). A pragmatic framework for single-site

and multisite data quality assessment in electronic

health record-based clinical research. Medical Care,

50:21–29.

Keuangan, D. P. K. (2012). Keputusan direktur jenderal

pajak no. kep-321/pj/2012.

Lee, S., Chen, T., Yu, L., and Lai, C. (2018). Image clas-

siﬁcation based on the boost convolutional neural net-

work. IEEE Access, 6:12755–12768.

Liu, S., Andrienko, G., Wu, Y., Cao, N., Jiang, L., Shi, C.,

Wang, Y., and Hong, S. (2018). Steering data qual-

ity with visual analytics: The complexity challenge.

Visual Informatics, 2:191–197.

Luo, Y., Zhang, T., Li, P., Liu, P., Sun, P., Dong, B., and

Ruan, G. (2020). Mdﬁ: Multi-cnn decision feature

integration for diagnosis of cervical precancerous le-

sions. IEEE Access, 8:29616–29626.

Martinez-Luengo, M., Shaﬁee, M., and Kolios, A. (2019).

Data management for structural integrity assessment

of offshore wind turbine support structures: data

cleansing and missing data imputation. Ocean Engi-

neering, 173:867–883.

Nikiforova, A. (2020). Deﬁnition and evaluation of data

quality: User-oriented data object-driven approach to

data quality assessment. Baltic Journal of Modern

Computing, 8:391–432.

Olah, C. (2015). Understanding lstm networks.

Ouyang, B., Song, Y., Li, Y., Sant, G., and Bauchy, M.

(2021). Ebod: An ensemble-based outlier detection

algorithm for noisy datasets. KnowledgeBased Sys-

tems, 231:107400.

Prokhorov, I. and Kolesnik, N. (2018). Development of a

master data consolidation system model (on the exam-

ple of the banking sector. Procedia Computer Science,

145:412–417.

Ridzuan, F. and Zainon, W. (2019). A review on data cleans-

ing methods for big data. Procedia Computer Science,

161:731–738.

Sadikin, M., Fanany, M., and Basaruddin, T. (2016). A new

data representation based on training data characteris-

tics to extract drug name entity in medical text.

Sadikin, M., Ramayanti, D., and Indrayanto, A. (2020). The

graded cnn technique to identify type of food as the

preliminary stages to handle the issues of image con-

tent abundant. In ACM International Conference Pro-

ceeding Series, page 108–113.

Sadikin, M., SK, P., and Bagaskara, L. (2021). The applica-

tion of machine learning approach to address the gpv

bias on pos transaction.

Sastrawan, I., Bayupati, I., and Arsa, D. (2022). Detection

of fake news using deep learning cnn–rnn based meth-

ods. ICT Express, 8:396–408.

Srivastava, S., Chaudhari, Y., Damania, Y., and Jadhav, P.

(2022).

Xuefeng Jiang, Yikun Wang, W. L. S. L. J. L. (2019). Com-

parative performance evaluation for image classiﬁca-

tion. International Journal of Machine Learning and

Computing, 9:10 18178 2019 9 6 881.

Zhang, Y. and Thorburn, P. (2022). Handling missing data

in near real-time environmental monitoring: A system

and a review of selected methods. Future Generation

Computer Systems, 128:63–72.

Comprehensive Approach to Assure the Quality of MSME Data in Indonesia: A Framework Proposal

161