International Journal of Machine Learning and Computing, Vol. 9, No.

4, August 2019

Applying Data Mining Techniques in Predicting Index and

non-Index Crimes
Allemar Jhone P. Delima

quite distinct [7], [8]. Another aim of this paper is to predict

Abstract—An increasing incidence of crime has led to the time, volume and or likelihood of future occurrence of index
development and use of computer-aided diagnosis system, tools and non-index crimes for 2018-2022 using ARIMA (1, 0, 7)
and methods in analyzing, classifying and predicting crimes. model. Predicting volume of crimes and the years of its high
This paper clusters municipalities in Surigao del Norte using
K-Means algorithm. This is instrumental in finding identical
occurrences is an essential strategy in alleviating future
traits, patterns and values in categorizing municipalities with problems before the adversaries can have the chance to
much, more, and most number of recorded index and non-index execute them. The police authority can identify and come up
crimes from 2013-2017. Prediction of its occurrence for the year with a strategic actions and preventive maintenance if the
2018-2022 was also provided using ARIMA(1,0,7) model. forecasted occurrence of crime and its possible volume of
Results showed that Surigao City has the most number of occurrences is identified.
recorded index and non-index crimes. Followed by the
municipality of Placer, Claver, and Dapa of cluster 2 and Del
TABLE I: ACTUAL AND FORECASTED DATA OF INDEX CRIMES FROM
Carmen of cluster 3. Further, physical injury, homicide,
2013-2017 AND 2018-2022 RESPECTIVELY IN THE PROVINCE OF SURIGAO
violation of special laws, car napping, reckless imprudence DEL NORTE
resulting to physical injury, and other non-index crimes has
26%, 25%, 25%, 24%, 24%, and 23% forecasted increase for
the year 2018-2022 with the highest occurrence in years 2018,
2018, 2020, 2018, 2020, and 2020 respectively. Future
researchers may utilize other data mining techniques supported
by a better accuracy result.

Index Terms—Crime forecasting, prediction, arima, data

mining, crime mining.

TABLE II: ACTUAL AND FORECASTED DATA OF NON-INDEX CRIMES FROM

I. INTRODUCTION 2014-2017 AND 2018-2022 RESPECTIVELY IN THE PROVINCE OF SURIGAO
The Police agencies around the world are anxious and DEL NORTE

eager to reduce crime [1]. Crime is a result of actions which

do not conform with the society’s norm and moral values [2],
condemned by the society for violation to any law. There is a
continuous increase in the analysis of crime in the past
decades hence, allowing many possibilities of studying and
extracting crime data from different disciplines through
different perspectives [3].
Data Mining, as the method of analyzing different type of
data to extract interesting patterns and knowledge [4], is also
used to discover critical information which can help local
authorities detect crime [5] as well as predicting areas which
have high probability for occurrence of crime and indicate
crime prone areas [6] including the type of crimes. II. RELATED LITERATURE
The goal of this paper is to create clusters to identify as to Data mining, as coined with the term Knowledge
what areas and municipalities within the province of Surigao Discovery in databases (KDD), used diverse approaches of
del Norte established the highest recorded index and non- DM analysis such as: decision tree, bayes classifiers,
index crimes from 2013-2017. This is done, by using association rules, clustering, neural networks, genetic
K-Means clustering algorithm. Data clustering is an algorithms, support vector machines, predicting or
unsupervised classification method used in creating groups of forecasting and many more. These are instrumental in
objects, or clusters, in such a way that objects in the same extracting important information from the database [9]. Along
cluster are very similar and objects in different clusters are with DM analytics, big data and predictive analysis are
relatively new concepts in criminology, while they have
Manuscript received December 29, 2018; revised June 11, 2018. become standard practice in disciplines such as business
A. J. Delima is with the College of Engineering and Information
intelligence, biomedical sciences, finance, and marketing.
Technology, Surigao State College of Technology, Surigao City, 8400,
Philippines (e-mail: [email protected]). The use of such techniques will also inadvertently impact

doi: 10.18178/ijmlc.2019.9.4.837 533

 International Journal of Machine Learning and Computing, Vol. 9, No. 4, August 2019

social sciences and humanities in general [10] and III. METHODOLOGY

criminology in particular [11] and is proved to be very useful The study was anchored on the concept of [19] but differs
as it is employed to perform in crime forecasting [12]. in many ways. Although K-Means algorithm was also used, it
The paper of [5] analyzed the crimes such as theft, was utilized to find clusters within the city and municipalities
homicide and various drug offences along with suspicious in the province of Surigao del Norte to identify areas with
activities, noise complaints and burglar alarm by using much, more, and most number of crimes recorded. The
qualitative and quantitative approach. Using K-means ARIMA (1,0,7) model was utilized to predict future
clustering data mining approach on a crime dataset from New occurrence of index and non-index crimes in the province
South Wales region of Australia, crime rates of each type of from 2018-2022.
crimes and cities with high crime rates have been found.
Moreover, a method to execute a problem-oriented fitness
function to predicted spatiotemporal patterns of criminal
activity, using memetic differential fuzzy approach which is a
unique time series approach from fuzzy clustering, in the city
of San Francisco, USA was presented in the paper of [3].
Results show that series approach of fuzzy clustering for
criminal patterns is a feasible method and is effective in
producing a forecast of criminal patterns.
Furthermore, [13] implemented data mining techniques to
understand certain trends and pattern of terrorist attacks in
India. K-means clustering was used to determine the year
wherein the terrorist groups were most active and also which
terrorist group has affected the most. The experimental result
is implemented in Rapidminer tool to determine the active
group and the affected year.
Fig. 1. Conceptual framework of the study.
As stated by [14], on the basis of data source, predicting
crimes can be categorized in to two strategies. One approach
is to use usual statistic techniques and models such as STL A. Clustering
[15], ARIMA, kernel density distribution, and more in Clustering is an unsupervised data analyzing technique
identifying hotspots of crime. Crime hotspots refers to used to divide the same data into the same group and the
geospatial locations where the probability for a crime to occur different data into the other group [21]. Index and non-index
is high. crime datasets of the city and municipalities from the province
Meanwhile, precise crime prediction for small areas like of Surigao del Norte for the year 2013-2017 were used as
police precincts is proposed in [16]. In [17], autoregressive presented in the Table III.
integrated moving average (ARIMA) is employed for near
future prediction of property crime. Based on 50 weeks TABLE III: INDEX AND NON-INDEX CRIME DATASET PER CITY AND
property crime data, an ARIMA model was built to predict MUNICIPALITY IN THE PROVINCE OF SURIGAO DEL NORTE
crime number of 1 week ahead. It is found that ARIMA model MUNICIPALITY INDEX CRIME NON-INDEX CRIME
ALEGRIA 151 313
has higher fitting and prediction precision than exponential BACUAG 148 297
smoothing. BURGOS 27 61
CLAVER 366 617
An approach based on auto-regressive models to reliably
DAPA 222 437
forecast crime trends in areas in Chicago was also performed. DEL CARMEN 151 216
In particular, ARIMA as a model to forecast the number of GENERAL LUNA 161 294
GIGAQUIT 183 264
crimes that is likely to occur in rolling time horizons was used MAINIT 194 303
to predict the number of crimes with an accuracy of 84% on MALIMONO 76 149
PILAR 98 138
one year-ahead forecasts and of 80% on two-year-ahead PLACER 316 801
forecasts [18]. SAN BENITO 65 121
In the Philippines, a recent study of [19] clustered the index SAN FRANCISCO 103 298
SAN ISIDRO 88 111
crime data in the province of Misamis Occidental using SISON 120 308
K-Means algorithm and forecasted its occurrence for the next STA. MONICA 85 185
SOCORRO 88 124
five years using ARIMA(1,0,1) model. Based on the result, 3 SURIGAO CITY 3562 3705
of the indexed crime data were categorized in the first cluster TAGANAAN 165 367
and the other five index crimes are found to be in the cluster 2. TUBOD 139 329

Moreover, rape, cattle rustling, physical injury, robbery, and

theft showed an increasing pattern based on the forecasted In K-means algorithm, the user specifies the k centroids.
data from 2015 to 2020. This K centroids refers to the number of the wanted clusters.
The study of [20] visualized forecasting algorithm Each cluster must have a centroid that is a mean of a cluster.
performance using time series instance space. They proved Then a nearest centroid is assigned to each data record. When
that the ARIMA method does the best overall forecasting all input data records have been assigned, the centroid
performance. changes in each cluster and is updated by calculating the mean

534
 International Journal of Machine Learning and Computing, Vol. 9, No. 4, August 2019

cluster. These processes will be repeated until the latest Fig. 3 shows that among the city and all municipalities in
centroids do not change [22]. Surigao del Norte, Surigao City has the highest crime rate
The experimental result for clustering was implemented with a total of 7,267 recorded index and non-index crimes
using KNIME (Konstanz Information Miner) [23] analytics from the year 2013-2017.
platform. Fig. 2 shows the node structure of K-means In Fig. 4, violation of special laws is the top recorded crime
clustering executed in KNIME. The node for the K-Means is in Surigao City. It is followed by other non-index crimes, theft,
connected and then positioned after the node of the imported physical injury, and robbery. The least recorded crimes in the
csv file of the dataset. The node color manager comes after as city are the cattle rustling, homicide, and reckless imprudence
it put distinctions to the results to be generated later. The node resulting to homicide.
scatter plot shows the scatter plot of the clusters while the
interactive table is used to view the result in a table manner

Fig. 2. Node structure of K-Means clustering in KNIME.

Fig. 4. Index and non-index crime rates in Surigao city.
Surigao City, as the only place that belongs to Cluster 1,
denotes that it does not possess any similar traits with the Municipalities that belongs to Cluster 2 is shown in Fig. 5.
other places. It is evident in the Table IV that having the The municipality of Placer has the highest recorded incident
highest value makes the algorithm categorized it differently of index and non-index crimes next to Surigao City of the
from the others. Municipalities under Cluster 2 is then Cluster 1. It is followed by the municipality of Claver and
interpreted as next to Surigao City to have the highest value of Dapa making them as the second, third and fourth most
recorded crimes. Meanwhile, municipalities that are under number of recorded crime incidents from the whole province
Cluster 3 has the least among the groups. of Surigao del Norte from the year 2013-2017.
In Cluster 3, the municipality of Del Carmen has the most
TABLE IV: CLUSTERING RESULT IN KNIME
number of recorded index and non-index crimes from the year
INDEX NON-INDEX
MUNICIPALITY CLUSTER 2013-2017 followed by Sta. Monica and Pilar. The
CRIME CRIME
ALEGRIA 151 313 2 municipality of Burgos has the lowest number of recorded
BACUAG 148 297 2
BURGOS 27 61 3
index and non-index crimes for the past five years as shown in
CLAVER 366 617 2 Fig. 6.
DAPA 222 437 2
DEL CARMEN 151 216 3
GENERAL LUNA 161 294 2
GIGAQUIT 183 264 2
MAINIT 194 303 2
MALIMONO 76 149 3
PILAR 98 138 3
PLACER 316 801 2
SAN BENITO 65 121 3
SAN FRANCISCO 103 298 2
SAN ISIDRO 88 111 3
SISON 120 308 2
STA. MONICA 85 185 3
SOCORRO 88 124 3
SURIGAO CITY 3562 3705 1
TAGANAAN 165 367 2 Fig. 5. Crime rates of each municipality from cluster 2.
TUBOD 139 329 2

Fig. 6. Crime rates of each municipality from cluster 3.

B. Forecasting Index Crimes

Fig. 3. Crime rates in Surigao del norte.
The study [19] used ARIMA(1,0,1) model in forecasting

535
 International Journal of Machine Learning and Computing, Vol. 9, No. 4, August 2019

index crimes, while the ARIMA(1,0,7) model was used in this from the actual homicide data from the past five years. It can
paper in determining the occurrence of index crimes for the be seen in Fig. 8 that there is a predicted increase of homicide
next five years. Fig. 7 to Fig. 14 showed the graph of the from year 2017 to 2018 but a decreasing trend from 2020 to
predicted index crimes from 2018 to 2020 having 80% and 2022.
95% interval. The total forecasted rape from 2018-2022 in the province
An autoregressive integrated moving average (ARIMA) of Surigao del Norte is 306 which is 5% higher from the actual
model makes prediction of time series values based upon rape data of year 2013-2017. It is forecasted that there is a
prior values (AR terms) as well as the errors made by previous considerable increase of rape cases from 2017 to 2018 and a
predictions (MA terms). This allows the model to adjust itself decrease years after. A slightly increase pattern from year
to sudden changes in the time series. Therefore, the ARIMA 2020 to 2022 is also evident in Fig. 10.
forecasting equation for a stationary time series is a linear
regression equation in which the predictors are the lags of the
dependent variable and/or lags of the prediction errors. This
model is explained in more detail in [18], [24]. In this paper,
this method was implemented in R Studio using R language.
It is shown in Fig. 7 that there is a decrease of recorded
murder cases in the province of Surigao del Norte since 2016
to 2017. The highest predicted rate of murder is 57 in the
years 2019 and 2022. The total predicted murder case for the
year 2018-2022 is 279 which is 4% lesser from the 290 actual Fig. 10. Forecasted rape from 2018-2022.
murder cases from the past five years.
It is evident in Fig. 9 that there is a rapid increase rate of It is presented in Fig. 11 that there is a decreasing trend of
physical injury from 2017 to 2018 with crime data of 608 robbery from year 2013 to year 2017. Meanwhile, the
which is also considered as the highest. A decrease is forecasted crime data of robbery shows a rapid increase from
forecasted in year 2018 to 2019 and an increase a year after. 62 actual data for the year 2017 to 146 forecasted data in year
The total forecasted crime data of physical injury from 2018 2018. An increasing and decreasing pattern is shown from
to 2022 in the province of Surigao del Norte is 2,508 which is 2017 to 2020 and 2020 to 2022, respectively. The total
26% higher from the past five years. forecasted crime data of robbery from 2018-2022 in the
province of Surigao del Norte is 1,203 which is 7% higher
from the past.

Fig. 7. Forecasted murder from 2018-2022.

Fig. 11. Forecasted robbery from 2018-2022.

The total forecasted theft from 2018-2022 in the province

of Surigao del Norte is 2,565 which is 4% higher from the
actual theft data of 2,461 from 2013-2017. An increasing
trend from 2017 to 2020 is evident in the Fig. 12.

Fig. 8. Forecasted homicide from 2018-2022.

Fig. 12. Forecasted theft from 2018-2022.

It is shown in Fig. 13 that there is a rapid increase from 23

Fig. 9. Forecasted physical injury from 2018-2022. recorded data of car napping in 2017 to forecasted data of 157
in the year 2018. A slightly decrease pattern is shown from
The total predicted homicide from 2018-2022 in the year 2018 all the way to year 2020. Meanwhile, an increasing
province of Surigao del Norte is 140 which is 25% higher pattern from 2020 to 2022 is also visible. The total forecasted

536
 International Journal of Machine Learning and Computing, Vol. 9, No. 4, August 2019

crime data of car napping from 2018-2022 in the province of The total predicted data for reckless imprudence resulting
Surigao del Norte is 492 which is 24% higher from the past to physical injury from 2018-2022 in the province of Surigao
five years. del Norte is 1,482 which is 24% higher from the 1,123 actual
data from the past five years. It can be seen in Fig. 16 that
there is a predicted increase from year 2017 to 2020 but a
decreasing trend from 2020 to 2022.

Fig. 13. Forecasted car napping from 2018-2022.

Among all the index crimes in the province of Surigao del

Norte, the cattle rustling has the lowest number of crime Fig. 16. Forecasted reckless imprudence resulting to physical injury from
2018-2022.
reported from the year 2013-2017. The forecasted crime data
of cattle rustling from 2018-2022 is 14 which has an estimated
It is shown in Fig. 17 that there is almost a steady trend
increase of 3 from the actual data of 11 from year 2013-2017.
from 153 recorded data of reckless imprudence resulting to
An increase from the year 2017 to 2018 and a decrease in the
damage to property in 2017 to forecasted data of 210 in the
next two years is shown in the Fig. 14. Further, a slightly
year 2018. A decrease and increase pattern is shown from
increase trend is evident in 2020-2022.
year 2018 to 2019 and 2019-2020 respectively and then
repeat all the way to year 2022. The total forecasted crime
data of reckless imprudence resulting to damage to property
from 2018-2022 in the province of Surigao del Norte is 958
which is 20% higher from the 760 actual data from
2014-2017.

Fig. 14. Forecasted cattle rustling from 2018-2022.

C. Forecasting Non-Index Crimes

ARIMA(1,0,6) model was used to forecast the occurrences
of non-index crimes for the next five years. Fig. 15 to Figure Fig. 17. Forecasted reckless imprudence resulting to damage to property
19 showed the graph of the predicted non-index crimes from from 2018-2022.
2018 to 2020 having 80% and 95% interval.
It is shown in Fig. 15 that there is an increase of recorded The predicted crime data for the violation of special laws
reckless imprudence resulting to homicide cases in the from 2018-2022 in the province of Surigao del Norte is
province of Surigao del Norte since 2014 to 2016 and a considerably high from the year 2017 to 2020 with forecasted
decreasing pattern thereafter. The highest predicted rate of 893 crime records. Meanwhile, there is a decreasing pattern
reckless imprudence resulting to homicide is 45 in the year from 2020 to 2022 as shown in Fig. 18. The total forecasted
2019. Further, a decreasing pattern is shown from year 2019 crime data of violation of special laws from 2018-2022 in the
to 2020 and a steady trend right after. The total predicted case province of Surigao del Norte is 3,959 which is 25% higher
for the year 2018-2022 is 209 which is 20% higher from the from the 2,974 actual data from 2014-2017.
166 actual cases from the past five years.

Fig. 18. Forecasted violation of special laws from 2018-2022.

Fig. 15. Forecasted reckless imprudence resulting to homicide from

2018-2022. It is presented in Fig. 19 that there is a decreasing trend of
non-index crimes from year 2014 to year 2017. Meanwhile,

537
 International Journal of Machine Learning and Computing, Vol. 9, No. 4, August 2019

the forecasted crime data shows a rapid increase from 256 [4] K. Rajalakshmi, S. S. Dhenakaran, and N. Roobini, “Comparative
analysis of K-means algorithm in disease prediction,” Int. J. Sci. Eng.
actual data for the year 2017 to 344 forecasted data in year Technol. Res., vol. 4, no. 7, pp. 2697–2699, 2015.
2018. An increasing and decreasing pattern is shown from [5] J. Agarwal, “Crime analysis using K-means clustering,” Int. J. Comput.
2017 to 2020 and 2020 to 2022, respectively. The total Appl. , vol. 83, no. 4, pp. 975–8887, 2013.
forecasted crime data of other non-index crimes from [6] O. Vaidya, S. Mitra, R. Kumbhar, S. Chavan, and R. Patil,
Comprehensive Comparative Analysis of Methods For Crime, pp.
2018-2022 in the province of Surigao del Norte is 2,530 715–718, 2018.
which is 23% higher from the 1,950 actual data from [7] D. Kaur and K. Jyoti, Enhancement in the Performance of K-means
2014-2017. Algorithm, vol. 2, no. 1, pp. 29–32, 2013.
[8] A. Bansal, M. Sharma, and S. Goel, “Improved K-mean clustering
algorithm for prediction analysis using classification technique in data
mining,” Int. J. Comput. Appl., vol. 157, no. 6, pp. 975–8887, 2017.
[9] E. Şuşnea, “Using data mining techniques in higher education,” High.
Educ., vol. 1, no. 1, pp. 68–72, 1996.
[10] R. Kitchin, “Big Data, new epistemologies and paradigm shifts,” Big
Data Soc., vol. 1, no. 1, p. 205395171452848, 2014.
[11] J. Chan and L. B. Moses, “Is big data challenging criminology?” Theor.
Criminol., vol. 20, no. 1, pp. 21–39, 2016.
[12] C. Yu, M. W. Ward, M. Morabito, and W. Ding, Crime Forecasting
Using Data Mining Techniques, 2011.
[13] P. Gupta, A. S. Sabitha, and T. Choudhury, Terrorist Attacks Analysis
Fig. 19. Forecasted other non-index crimes from 2018-2022. Using Clustering Algorithm, © Springer Nat. Singapore Pte Ltd., pp.
317–328, 2018.
[14] J. Azeez and D. J. Aravindhar, “Hybrid approach to crime prediction
using deep learning,” in Proc. 2015 Int. Conf. Adv. Comput. Commun.
IV. CONCLUSION Informatics, 2015, pp. 1701–1710.
With the use of K-Means clustering algorithm, determining [15] A. Malik, R. Maciejewski, S. Towers, S. Mccullough, and D. S. Ebert,
“Proactive spatiotemporal resource allocation and predictive visual
the groupings of municipality with identical traits and values analytics for community policing and law enforcement,” IEEE Trans
became possible. In cluster 1, the Surigao City topped as the Vis Comput Graph., vol. 20, no. 1, pp. 1863–1872, 2014.
municipality in Surigao del Norte with most number of [16] W. Gorr, A. Olligschlaeger, and Y. Thompson, “Short-term forecasting
of crime,” Int. J. Forecast., vol. 19, no. 4, pp. 579–594, 2003.
reported index and non-index crimes. In cluster 2, the [17] P. Chen, H. Yuan, and X. Shu, “Forecasting crime using the ARIMA
municipality of Placer, Claver and Dapa has the highest crime model,” in Proc. 5th Int. Conf. Fuzzy Syst. Knowl. Discov. FSKD,
rate. Meanwhile in cluster 3, Del Carmen, Sta. Monica and 2008, vol. 5, pp. 627–630.
[18] E. Cesario, C. Catlett, and D. Talia, “Forecasting crimes using
Pilar was identified. autoregressive models,” in Proc. 2016 IEEE 14th Int. Conf.
Among the index crimes in the province of Surigao del Dependable, Auton. Secur. Comput., pp. 795–802, 2016.
Norte, theft was identified as the highest number of recorded [19] M. Y. Orong, A. M. Sison, and A. A. Hernandez, “Mitigating
vulnerabilities through forecasting and crime trend analysis,” in Proc.
crime with a total of 2,565 from 2013-2017 with the highest
2018 5th Int. Conf. Bus. Ind. Res., 2018, pp. 57–62.
occurrence in 2014. Furthermore, the highest predicted crime [20] Y. Kang, R. J. Hyndman, and K. Smith-Miles, “Visualising forecasting
for the year 2018-2022 is the physical injury having the algorithm performance using time series instance spaces,” Int. J.
predicted value of 2,508 or 26% increase from 2014-2017. Forecast., vol. 33, no. 2, pp. 345–358, 2017.
[21] A. Ben Ayed, M. Ben Halima, and A. M. Alimi, “Survey on clustering
Moreover, the least reported crime in the province is cattle methods: Towards fuzzy clustering for big data,” in Proc. 6th Int. Conf.
rustling. Soft Comput. Pattern Recognition, 2015, pp. 331–336.
For the non-index crimes, violation of special laws was [22] A. Thammano and A. K. Algorithm, Enhancing K-means Algorithm
for Solving Classification Problems, pp. 1652–1656, 2013.
identified as the highest reported incident in the province with [23] L. Feltrin, “KNIME an open source solution for predictive analytics in
the highest occurrence in 2014. Moreover, violation of special the geosciences [software and data sets],” IEEE Geosci. Remote Sens.
laws has the highest predicted value of 3,959 or 25% increase Mag., vol. 3, no. 4, 2015.
[24] A. Rege et al., “Predicting adversarial cyber intrusion stages using
from the data of year 2014-2017 with the highest occurrence autoregressive neural networks,” IEEE Intell. Syst., 2018.
in 2020.

REFERENCES Allemar Jhone P. Delima was born in Surigao

City, Surigao del Norte, Philippines on March 24,
[1] G. Dudfield, C. Angel, L. W. Sherman, and S. Torrence, “The ‘power 1994. He completed master in information
curve’ of victim harm: targeting the distribution of crime harm index technology degree at Surigao State College of
values across all victims and repeat victims over 1 year,” Cambridge J. Technology, Surigao City, Philippines in 2016.
Evidence-Based Polic., vol. 1, no. 1, pp. 38–58, 2017. With his quest in improving his professional
[2] M. Sevri, H. Karacan, and M. A. Akcayol, “Crime analysis based on career, he took doctoral degree and is now a student
association rules using apriori algorithm,” International Journal of of doctor in information technology at
Information and Electronics Engineering, vol. 7, no. 3, 2017. Technological Institute of the Philippines, Quezon City, Philippines. He is
[3] C. D. R. Rodriguez, D. M. Gomez, and M. A. M. Rey, “Forecasting currently connected in Surigao State College of Technology as a faculty in
time series from clustering by a memetic differential fuzzy approach: the College of Engineering and Information Technology. His research
An application to crime prediction,” in Proc. 2017 IEEE Symp. Ser. interest is in data mining and data analytics.
Comput. Intell., 2018, pp. 1–8.

538