See discussions, stats, and author profiles for this publication at: https://www.researchgate.

net/publication/324919218

Accidental Deaths in India : Forecasting with ARIMA Model

Article · May 2018

CITATIONS READS

0 39

4 authors, including:

Arnab N. Patowary Manash Pratim Barman

Assam Agricultural University Dibrugarh University
18 PUBLICATIONS   9 CITATIONS    19 PUBLICATIONS   71 CITATIONS   

SEE PROFILE SEE PROFILE

Srinivasa Rao Gadde

Dodoma University
92 PUBLICATIONS   600 CITATIONS   

SEE PROFILE

Some of the authors of this publication are also working on these related projects:

Control charts View project

Bayesian analysis of time series models View project

All content following this page was uploaded by Arnab N. Patowary on 07 May 2018.

The user has requested enhancement of the downloaded file.

 761

Environment and Ecology 36 (3) : 761—766, July—September 2018

Website: environmentandecology.com ISSN 0970-0420

Accidental Deaths in India : Forecasting with ARIMA Model

A. N. Patowary, B. Dutta, M. P. Barman, S. R. Gadde

Received 9 March 2018; Accepted 29 March 2018; Published on 18 April 2018

Abstract Accident is one of the burning problems

Introduction
for pre-mature end to human lives. Road accident in
India is an increasing trouble and has raised one of
An accident is an unexpected incident that results in
the country’s major problems. This paper outlines
injury, deaths, damage to property or some other
development of a conventional time series model viz.
losses. It is one of the vital reasons for pre-mature
Autoregressive Integrated Moving Average
end to human lives. Injuries resulting from the acci-
(ARIMA) model for the annual total number of deaths
dent make many people handicap and these adverse
due to accident (natural and unnatural) in India cov-
victims are important members of a family as well as
ering the period 1967 to 2015 and to forecast the num-
the society. Finally, the lost due to accident is felt by
ber of annual accidental deaths likely to occur in fu-
a family and the society. Accidental deaths and inju-
ture. We investigated and found that ARIMA (2, 2, 1)
ries influence the economic development of a coun-
model is suitable for the given data set.
try as major portion of the victims are economically
active which support the family for their livelihood.
Keywords ARIMA, Forecasting, India, Accidental
National Crime Record Bureau classified causes of
death.
accidental deaths into three categories viz. natural
accidents, un-natural accidents and other causes of
accidental death (sudden deaths with no apparent
cause of death like heart attacks, death during preg-
nancy, poisoning,. Due to the advantages of modern
technology and medical discoveries, the proportion
A. N. Patowary* of accidental deaths due to natural causes (exposure
College of Fisheries, Assam Agricultural University,
to cold, starvation, epidemic, cyclone) is seem to be
Raha 782103, Assam, India
e-mail: [email protected] in decreasing trend. On the other hand, incidence of
road accident is increases day by day. In India, every
B. Dutta, M. P. Barman one hour nearly 1 person died due to cause attribut-
Department of Statistics, Dibrugarh University,
able to force of nature while 38 persons killed in un-
Dibrugarh 786004, Dibrugarh, India
e-mail: [email protected] natural accidents in the year 2015. Out of 4,13, 457
[email protected] accidental deaths, about 81.3% deaths are due to un-
natural causes and 16.2% deaths are due to other
S. R. Gadde
causes and the remaining 2.5% are due to causes at-
Department of Statistics, Dodoma University, Tanzania
e-mail: [email protected] tributable to force of nature 2015 (National Crime
*Corresponding author Record Bureau 2015). Most of the victims of un-natu-
 762

ral accidental deaths are in the age group of 18 to 45 Ghana as well as makes a five-year forecast. The re-
years, accounting for 61.6% of total persons killed in sults reveal that ARIMA (0, 2, 1) is the best model to
such accidents in India during the year 2015 (National analyze the road traffic accident in Ghana. Similarly,
Crime Record Bureau 2015). Adebola et al. (2015) used ARIMA model of time se-
ries, to analyzed yearly data on the consequences of
road accident data in Nigeria from 1960-2013. Further,
Road Traffic Accident (RTA) which is identified Mutangi (2015) used ARIMA model for the annual
as the third most important cause of overall mortality Zimbabwe traffic accident statistics from 1997 to 2013
and the main cause of death among the age group 15- and to forecast the number of annual traffic accidents
44 year and it represents 12% of global burden of likely to occur in future.
disease (Sivakumar and Krishnaraj 2015). High rate of
road accidents and the resulted number of deaths
and injuries is a major concern for India which ad- With this background, an attempt has been made
versely affects the social structure and economy. Con- to know the pattern of accidental deaths in India from
sidering the gravity of the situation, it is necessary to 1967-2015 and to forecast the number of annual acci-
take appropriate measures for reducing this high level dents likely to occurs in future using appropriate
of deaths and injuries. Forecasting of accidents or ARIMA model. Previous study showed that if mean
number of deaths due to it may be very valuable in- count of a data set is very high, then error may be
formation for taking different safety measures. Fore- approximated as normal and ARIMA model may be
casting help the policy makers as well as engineers to satisfactory (Quddus 2008). For this reason, we have
investigate the safety system to prevent accidents used ARIMA model instead of using other time se-
and number of deaths due to it in the up-coming ries models for count data. The organizations of the
days. paper are as follows: Section 2 deals with source of
the data and complete methodology undertaken in
this study. In Section 3, we have given results and
Time series is a sequence of numerical data that discussion of the study. Finally, conclusion of the
generally occurs in uniform intervals over a period of study is given in Section 4.
time. The main application of time series analysis is
forecasting by analyzed historical data (Monfared et
al. 2013). Forecasting techniques is usually applied Materials and Methods
as an aid in controlling past and present operations
in planning for future needs. Among the most effec-
tive approaches for analyzing time series data is the The data are secondary in nature and obtained from
ARIMA (Autoregressive Integrated Moving Aver- National Crime Record Bureau, Ministry of Home Af-
age) model introduced earliers. fairs annual Report on Accidental Deaths and Sui-
cides in India covering the period from 1967 to 2015.

There are extensive literatures available of ap- The methodology and the theorems propounded
plying ARIMA model in accidental data (mainly road earlier called the Autoregressive Integrated Moving
accidents) in different countries of the world includ- Average (ARIMA) has been used . This is an ad-
ing India. Ofori et al. (2012) used Exponential Smooth- vance technique of forecasting requires long seasonal
ing and ARIMA model to forecast road accident inju- time series data. This model decomposes historical
ries in Ghana. Also, Quddus (2008) used a time series data into an Autoregressive (AR) process, where there
model to understand the trend of accidents and as- is a memory of past values, an Integrated (1) process,
certain the viability of applying ARIMA models on which accounts for stabilizing or making the data sta-
data from Taybad city. Moreover, Avuglah et al. (2014) tionary plus a Moving-Average (MA) process, which
applies the ARIMA time series model to study the accounts for previous error terms making it easier to
trends and patterns of road traffic accident cases in forecast.
 763

Table 1. Parameter estimation of ARIMA (2, 2, 1)

Parameter Estimate Std. error Z p-value

^
1 -0.5 277 0.1679 -3.1 421 0.00001
^
 -0.3 560 0.1809 -1.9 684 0.00001
^
1 -0.8 017 0.1060 -7.5 638 0.00001

differenced series is still behave non stationary (Fig.

2). Therefore, for better result, the researchers go for
second differencing of the data (Fig. 3). A clear non
Fig. 1. Plotting of original data.
stationary behavior of the second differenced data
has been observed in (Fig. 4). Similarly, Augmented
Dickey-Fuller test tells the same truth (p-value <0.05).

Results and Discussion In Fig. 4, it is observed that there is single signifi-

cant spike at lag 1 in the ACF indicating MA (1) pro-
Fig. 1 shows the time plot of total number of deaths cess whereas two significant spikes have been ob-
due to accident in India from 1967-2015. From Fig. 1, served at lag 1 and lag 2 in the PACF suggesting AR
it is observed that the time plot exhibits a significant (2) process might be feasible. Therefore, the researcher
upward trend. Next, we divide the whole data set into acquire a rough idea that ARIMA (2, 2, 1) model, may
two parts: testing (1967-2005) and validation (2006- be tentatively appropriate to total accidental death
2015). First, we develop the ARIMA model for testing data which is given by
part and compare the forecasted values from the model
with validation part. If the model fits well and fulfills  (1–1 B–2B2) (1–B)2Yt =(1– 1B) et (3.1)
the all assumptions, then we refit the model for the
whole data set i.e. 1967-2015 and forecast accidental
deaths for the upcoming 10 years. Autocorrelation For the ARIMA (2, 2, 1) model obtained above
Function (ACF) and Partial Autocorrelation Function Eq. (3.1), the parameters of the model are estimated
(PACF) of testing data set reveal the significant trend. by maximum likelihood method obtained from R soft-
ware is given in Table 1. In Table 1, it is observed that
By applying augmented Dickey-Fuller test to the all the coefficients are significant. The constant term
first differenced data, it is found that p-value >0.05, of the model is omitted due to differencing. Then the
so we cannot reject the null hypothesis i.e. the first extimated model can be written as

Fig. 2. Firstdifference of the original data.data. Fig. 3. Second difference of the original data
 764

Fig. 4. ACF and PACF of the second difference of the original data.

1+0.5277×B+0.35602 ×B 2)(1–B) 2 Y t=(1+0.8017×B)et (3.2) that p value is 0.4711 which means that the residuals
series is white noise (with mean 0 and variance 2). To
If the model ARIMA (2, 2, 1) fits well, the stan- be sure that the predictive model cannot be improved
dardized residuals estimated from this model should upon, it is also a good idea to check whether the
behave as an i.i.d.(independent and identically dis- forecast errors are normally distributed or not, quintile-
tributed) sequence with mean zero and variance 2. quintile (q-q) plot of the residuals is drawn which is
From standardized plot of residuals, it is observed given in (Fig. 6).
that residuals are lie within the limit of -3 and +3. Also,
(Fig. 5) shows, none of the autocorrelations is indi- Moreover, by applying Komogorov-Smirnov (K-
vidually statistically significant and nor the Ljung- S) test to the residuals it is found that D=0.0277 and
Box-Pierce Q-statistics are statistically significant. p-value =0.3212 which indicates that the residuals fol-
Here, the Ljung-Box test statistic is 12.22 and p-value low normal distribution well. Now, to check the valid-
is 0.2011, so the null hypothesis of independence of ity of the fitted model, the actual observations are
the residual series cannot be rejected. Using the white plotted with predicted values from 2006-2015 (10
noise test (from the normwn.test package in R : Per- years) in (Fig. 7) where blue line represents predicted
form a univariate test for white noise), it is obtained numbers of death from the model and red line repre-

Fig. 5. Standardized residuals, the ACFof the residuals and the p-values of the Q-statistic.
 765

Fig. 6. Q-Q plot of the residuals. Fig. 8. Trend of accident cases in India for upcoming 10
years.

sents actual observations. From Fig. 7, it is observed Table 2. Forecasting accident cases in India for upcoming 10
that the total number of predicted deaths due to acci- years.
dents in India from 2006-2015 is almost equal and ex- 95% CI
act pattern with the actual data. Therefore, it may be Year Point forecast Lower limit Upper limit
concluded that proposed model would be good fitted 20 16 436 031 408 836 463 225
to the yearly total numbers of death in India due to 20 17 458 363 425 619 491 106
accident from 1967-2015. 20 18 459 150 421 455 496 844
20 19 471 391 425 009 517 773
20 20 485 259 431 155 539 363
20 21 494 190 432 405 555 976
Moreover, the forecasted number of deaths due 20 22 505 148 434 678 575 617
to accidents in India for the upcoming 10 years which 20 23 516 793 437 461 596 126
is given in Table 2 along with 95% confidence inter- 20 14 527 355 438 935 615 774
20 25 538 243 440 289 636 197
val. Also, (Fig. 8) represents forecasted values along
with 99% and 95% Confidence Interval (CI).
Conclusion

In this paper, the yearly deaths due to accident in

India have been studied using the Box-Jenkins
(ARIMA) methodology. The estimation and diag-
nostic analysis results revealed that the models’ are
adequately fitted to the historical data. The residual
analysis, confirmed that there is no violation of as-
sumptions in relation to model adequacy. Therefore,
it may be concluded that, selected model ARIMA (2,
2, 1) would be good fitted to the yearly total number
of deaths in India due to accident from 1967-2015.
The forecasted number of deaths due to accidents in
India for the upcoming 10 years also exhibits an up-
ward trend. Therefore, the policy makers and manag-
ers should pay more attention to prevent accidents in
Fig. 7. Plotting of predicted values with actual observations
(1976-2015). the up-coming days. Most of the accidental deaths
are un-natural and human error is one of the major
 766

causes of this problem. To avoid increase in un-natu- road traffic accident in Nigeria: Time series approach.
ral accident, the government should create awareness Int J Comp Appl Technol and Res 4 (4) : 262—273.
Avuglah RK, Adu-Poku KA, Harris E (2014) Application of
program among the masses. For example, to prevent ARIMA models to road traffic accident cases in Ghana.
road accident; Road Safety Agency should create Int J Stat and Appl 4 (5) : 233—239.
awareness program among the masses so as to notify Monfared AB, Soori H, Mehrabi Y, Hatami H, Delpisheh A
(or remind) them of the rules and regulations of trans- (2013) Prediction of Fatal road traffic crashes in Iran
using the Box-Jenkins time series model. J Asian Sci
portation. Also, the law makers should give severe Res 3 (4) : 425—430.
and immediate penalty for the law breakers. Un-natu- Mutangi K (2015) Time series analysis of road traffic acci-
ral accidents are control of human hand, only precau- dents in Zimbabwe. Int J Stat and Appl 5 (4) : 141—149.
tions are necessary. The entire nation must aware of National Crime Record Bureau, Ministry of Home Affairs
annual report on accidental deaths and suicides in India
this burning problem and should obey the rules and covering the period from 1967 to 2015.
regulations. Natural accidents are beyond the con- Ofori T, Ackah B, Ephraim L (2012) Statistical models for
trol of human hand but can take preventive measures forecasting road accident injuries in Ghana. Int J Res
on time. Furthermore, government as well as masses in Environ Sci and Technol 2 (4) : 143—149.
Quddus MA (2008) Time series count data models : An em-
should not be left alone to solve these burning prob- pirical application to traffic accidents. Accident Analy-
lems. sis and Prevention 40 (5) : 1732—1741.
Sivakumar T, Krishnaraj R (2015) Relationship between road
Re fe r e nc e s traffic accidents and socio-demographic factors in In-
dia. Int J Pharmaceut Sci Review and Res 32 (2) : 137—
Adebola FB, Sansi RA, Adegoke NA (2015) Consequence of 14 4.

View publication stats