Madras Agric. J., 2021; https://doi.org/10.29321/MAJ.10.

000546

RESEARCH ARTICLE
Use of Statistical Models in Predicting Groundnut Yield in Relation to
Weather Parameters
Abhinaya D1*, Patil S G1, Dheebakaran Ga2, Djanaguiraman M3, Arockia Stephen Raj1
1*
Department of Physical Sciences & Information Technology, Agricultural Engineering College & Research Institute,
Tamil Nadu Agricultural University, Coimbatore-641 003.
2
Agro Climate Research Centre, Tamil Nadu Agricultural University, Coimbatore.
3
Department of Crop Physiology, Agricultural College & Research Institute, Tamil Nadu Agricultural University, Coimbatore.

ABSTRACT

In Tamil Nadu, groundnut is an essential and major oilseed crop, mainly

grown under rainfed conditions. The changes in weather parameters might
affect the productivity of groundnut. Hence, crop yield forecasting based
on weather parameters is essential for proper planning, decision-making,
and buffer stocking policy formulation. As for the data with multicollinearity,
penalized regression models i.e.Ridge, Least Absolute Selection and
Shrinkage Operator (LASSO) and Elastic Net (ENet), are better alternatives
to classical linear regression. The data on weather parameters such as
maximum temperature(Tmax), minimum temperature (Tmin), morning
relative humidity (RH I), evening relative humidity (RH II), and rainfall were
collected for 29 years from 1991-2019. The weather indices approach was
used in this study. The collected data were partitioned into training, and
testing datasets and the hyperparameters of penalized regression models
were tuned using cross-validation. The performance of the models was
evaluated using an adjusted coefficient of determination (R2adj), Root Mean
Squared Error (RMSE), normalized RMSE (nRMSE), Mean Absolute Error
(MAE) and Mean Absolute Percentage Error (MAPE) as the goodness-of-fit
criteria. The results revealed that all the Penalized regression models provide
a better fit to data. The SMLR and ENet were found to predict with better
accuracy. Hence, these methods can be used for groundnut yield forecasting
during Kharif season for the Coimbatore district of Tamil Nadu.

Keywords: Stepwise Multiple Linear Regression; Ridge regression; LASSO; Elastic Net;
Groundnut yield prediction, Weather indices.

INTRODUCTION Crop yield is influenced by technological change

and weather variability. Technological factors
Groundnut is one of the most essential oilseed
increase yield smoothly through time and therefore,
crops of India. In Tamil Nadu, it is annually grown in
years or some other parameters of time can be
about 0.35 million hectares with about 0.98 million
used to study the overall effect of technology on
tons production. In India, Tamil Nadu ranks third in
yield (Agarwal et al., 1980). Generally, there are
production contributing 9.74% of the total production
two approaches for crop yield forecasting: crop
of groundnut crop in the country, with an average
simulation and empirical statistical models (Bocca
yield of 2840 kg ha-1 (Directorate of Economics
and Rodrigues,2016). Though crop simulation
and Statistics, 2019-20). In Tamil Nadu, two-thirds
models are precise, they are input data-intensive
of the groundnut cultivation area is under rainfed
and the lack of sufficient data sets makes their
condition and the remaining one-third is under
application limited to smaller scales rather than
irrigated condition.
regional scales.
The overall growth of the Indian economy relies
Hence, empirical statistical models with simple
on the performance of agriculture, which depends ,
regression techniques have been largely used
upon the weather conditions every year. A timely and
as an alternative to process-based simulation
reliable forecast of crop yield is of great importance
models (Lobell and Burke,2010; Shi et al ,2013).
for monsoon-dependent countries like India,
Calibrated and tested statistical models lead to
where the economy is mainly based on agricultural
successful crop yield forecasting based on weather
production (Vinaya et al., 2017).
Corresponding author mail id: abhinayaduraisamy@gmail.com 107 | 10-12 | 1
parameters. Since groundnut is mainly cultivated regression model has been applied to detrend the
under rainfed conditions, weather conditions yield of groundnut.
highly affect its productivity. Most of the previous
Yt= β0 + β1*t (1)
studies have used Multiple Linear Regressions
(MLRs) to develop a statistical crop yield prediction Where, t is the time period, Yt is the crop yield
model (Rai et al, 2013; Dhekle et al,2014; Kumar at time t, and β0 and β1 are the coefficients. This
et al,2014). However, MLR results in over-fitting model’s residuals (detrended yield) were used for
when (a) the number of samples is less than the indices calculation (Trnka et al.,2009).
number of predictors, (b) multicollinearity exists
Weather Indices Approach
i.e. when the independent variables are correlated
(Verma et al., 2016). One of the consequences For each weather variable, two indices are
of multicollinearity is the large standard errors of developed, one as the total values of weather
regression coefficients, making the inaccurate parameters over different weeks and the other one
inference based on the fitted model (Yakubu, 2010; as weighed total; the weights are the correlation
Dormann et al., 2013). coefficient between detrended yield and weather
variable in respective weeks.
To overcome these drawbacks, feature selection m
and penalized regression methods such as Stepwise Unweighed weather indices: Zij = ∑w=1 Xiw
Multiple Linear Regression (SMLR), least absolute m j

shrinkage and selection operator (LASSO), elastic Weighed weather indices: Zij = ∑w=1 riw Xiw
net (ENET) and ridge regression techniques can Where,
be used (Das et al.,2017). In this context, the main
Xiw = value of ith weather variable in w-th week
objective of our study is to develop and select a
j
statistical groundnut yield forecasting model for the riw = correlation coefficient of detrended yield with
Coimbatore district of Tamil Nadu with the predictive i-th weather variable
performance and efficiency of the developed models.
m = week of forecast
MATERIAL AND METHODS For j=0, we have unweighed indices and for j=1,
Data Collection weighed indices. Totally 11 weather variables were
generated as per the procedure mentioned above
Time series data of groundnut yield (Arachis
is presented in Table.1.
hypogea L) for Coimbatore district of Tamil Nadu
for 29 years (1991 to 2019) has been collected YIELD FORECAST MODELS
from the Season and Crop Report, Department of Stepwise Multiple Linear Regression
Economics and Statistics. Daily weather data were
collected from Agro Climate Research Centre, TNAU. Multiple Linear Regression (MLR) is the most
The data on five weather variables namely maximum straightforward approach for the development of
temperature (Tmax, oC), minimum temperature statistical models. However, its application for the
(Tmin,oC), morning and evening relative humidity dataset with more significant explanatory variables
(RH I & RH II (%)) and rainfall (mm) for a total of 18 and is not always successful (Balabin et al., 2011).
weeks of crop cultivation which includes 14 th to 31 A stepwise regression procedure was adopted
st
standard meteorological week (SMW) has been to select the best regression variables among
used in the study as the sowing of the groundnut many independent variables (Singh et al, 2014).
in Coimbatore district is usually carried out during A fundamental problem with stepwise regression
the month of April-May (Chithiraipattam). Daily data is that some real explanatory variables that have
of Tmax, Tmin, RH I and RH II had been converted causal effects on dependant variables may happen
into its weekly average, whereas the weekly sum to be statistically insignificant, while nuisance
of rainfall has been considered. Out of the 29-year variables may be coincidentally significant (Smith
data, 24 years were used for calibration, while the et al., 2018). Hence, we opt for alternative methods
remaining 5 years were used for validation. such as penalized regression methods.
Detrending of Yield Time Series Data Penalized Regression
The fluctuations in yield data over the years due Penalized regression is a better alternative
to the technology differences, climatic variability, for the linear regression model (or the ordinary
etc., leads to a nonlinear and non-stationary trend least squares method). The penalized regression
which has to be removed. The correlation between adds a constraint (penalty) in the equation. The
detrended yield and weather parameters is used to consequence of imposing this penalty is to reduce
calculate weight for model development(Wu et al., the coefficient values towards zero. This allows the
2007). In the present investigation, a simple linear less contributive variables to have a coefficient close

107 | 10-12 | 2
to zero or equal to zero. The logic behind penalized cross-validation (Piaskowski et al., 2016). The
regression is to reduce the impact of multicollinearity overall strength of the penalty is controlled by tuning
since all independent variables in the study are parameter λ (Hastie and Qian,2014). The other
related. tuning parameter alpha was set at 0 for Ridge, 1 for
LASSO and 0.5 for ELNET. The data were analyzed
Ridge Regression
using ‘glmnet’ R-package (Friedman et al., 2009).
Ridge regression shrinks the regression
Model Performance
coefficients so that variables with a minor
contribution to the outcome have their coefficients The performance of the developed statistical
close to zero. The shrinkage of the coefficients is models is tested using, adjusted R2, root mean
achieved by penalizing the regression model with square error (RMSE), normalised RMSE, mean
a penalty term called L2-norm, which is the sum of absolute error (MAE) and mean absolute percentage
the squared coefficients (Zou and Hastie, 2005). error (MAPE) were calculated using the following
formula:
L2 = ∑ (Ŷi– Yi)² + λ∑ β² (2)
RMSE = [1/n ∑(i=1) (Yi– Ŷi)²]
1/2
n
(5)
Where y is the independent variable, 𝛽 is the
corresponding coefficient and λ is the L2 norm nRMSE = [1/n ∑(i=1) (Yi– Ŷi)²]
1/2
n
X 100/(mean(yi)) (6)
penalty. A large value of λ means a more significant
MAE = 1/n ∑ (i=1) |Yi– Ŷi|
n
(7)
amount of shrinkage. Ridge regression keeps all the
predictors in the model without making any variable MAPE = 100/n ∑ (i=1)|(Yi– Ŷi) Yi |
n
(8)
selection.
yi = actual value
Lasso Regression (Least Absolute Shrinkage And
Ŷi = Model output
Selection Operator)
R2adj towards 1 and RMSE towards 0 indicates
It shrinks the regression coefficients toward
better performance of the developed models.
zero by penalizing the regression model with a
Also lesser the MAE and MAPE values, the better
penalty term called L1-norm, which is the sum of the
fit the model is. According to nRMSE, the model
absolute coefficients. In the case of lasso regression,
performance is judged as excellent, good, fair and
the penalty has the effect of forcing some of the
poor when the values are in the range of <10%, 10–
coefficient estimates, with a minor contribution to
20%, 20–30% and >30%, respectively (Jamieson
the model, to be exactly equal to zero (Tibshirani,
et al., 1991).
1996). One obvious advantage of lasso regression
over ridge regression is that it produces more RESULTS AND DISCUSSION
straightforward and more interpretable models that
Summary Statistics Of Yield Data
incorporate only a reduced set of predictors.
The summary statistics of groundnut yield data
L1= ∑(Ŷi– Yi)² + λ∑ |β| (3)
(1991-2019) of the Coimbatore district of Tamil
Where y is the independent variable, 𝛽 is the Nadu is presented in Table 2. The maximum yield
corresponding coefficient and λ is the L1 norm was 2877 kgha-1, whereas the minimum yield was
penalty. 1519 kg ha-1. The coefficient of variation of yield
was found to be 17.84%. A normal Q-Q plot was
Elastic Net Regression
constructed for testing the normality of yield data
Elastic Net combines characteristics of both and it was affirming the normality, thus satisfying
lasso and ridge, i.e., penalized with both the L1 and L2 the basic assumptions of parametric models
norm (Hoerl and Kennard, 1970). The consequence (Fig.1). Figure.2. Shows the Pearson’s coefficient of
of this is to effectively shrink coefficients (like in ridge correlation between all variables. Significant positive
regression) and to set some coefficients to zero (like correlations (correlation coefficient greater than
in LASSO). Hence it reduces the impact of different 0.5) were found between yield and Z11, Z30, Z31,
features while not eliminating all of the features
Table1. Weather indices used in the development of
(Cho et al., 2009).
multivariate yield forecasting regression
L = ∑(Ŷi– Yi)² + λ2∑ β² + λ1∑ |β| (4) models
Where y is the independent variable, 𝛽 is the Parameter Unweighed Indices Weighed Indices
corresponding coefficient and λ is the penalty. Tmax Z10 Z11
Tmin Z20 Z21
These methods have two parameters, namely
lambda and alpha, which need to be optimized. The Rain Z30 Z31
optimal lambda values were selected by minimizing RH I Z40 Z41
the average mean square error in leave-one-out RH II Z50 Z51

107 | 10-12 | 3
Z51 (P<0.05). The yield was found to be strongly opted for alternative approaches to fix this problem
correlated with those variables (p<0.01) and the of multicollinearity.
correlation coefficients ranged between 0.50 and Figure.1. Normal Q-Q plot for groundnut yield of
0.59 (Iqbal et al., 2019). Coimbatore district.
Table.2. OLS estimates of regression coefficients
in MLR and VIF
Variable Coefficients VIF
Intercept 5249.95 ± 5597.78 0
Z10 -14.37 ± 11.0972 12.01887
Z11 28.88 ± 38.93 7.859344
Z20 6.62 ± 7.73 3.753992
Z21 -21.24 ± 41.89 6.089400
Z30 -23.48 ± 16.18 474.9300
Z31 57.18 ± 36.39 507.5759
Z40 -0.39 ± 0.91 6.801643
Z41 4.58 ± 5.49 5.729535
Z50 -1.01 ± 0.86 4.468881
Z51 13.30 ± 3.41 1.651192
T 9.71 ± 8.84 3.438897
Adj R2 0.8215
RMSE 184.09

Groundnut Yield Forecasting Models

In Multiple Linear Regression (MLR), regression This study utilized Ridge, LASSO, ENet regression
coefficients along with their standard errors and and SMLR as alternative methods to MLR due to the
VIF values are shown in Table.3. All the predictors presence of multicollinearity. In the regularization
included in the model explained 82.15% of the techniques such as ridge, LASSO and ENet cross-
variation. The VIF values of more than 5 observed for validation is done for selecting the optimal lambda
most of the variables may be considered a cause of (λ min) values (Fig.3). The results of applying these
concern, whereas a value of more than 10 indicates methods for groundnut yield prediction are shown
severe multicollinearity (Sheater, 2009; Kutner et al., in Table.4.
2004). As the data needed further examination, we

Table.3. Descriptive statistics of variables

Rainfall
Tmax (0C) Tmin (0C) RH-I (%) RH-II (%) Yield (kg ha-1)
(mm year-1)
Mean 33.13 23.69 86.09 41.24 1194 2045

Maximum 38.10 26.20 92.43 59.71 1303 2877

Minimum 25.83 19.53 78.86 21.86 1097 1519

Standard deviation 1.44 1.34 2.88 7.59 46.15 365

CV (%) 4.34 5.65 3.35 18.40 3.86 17.84

Stepwise Multiple Linear Regression (SMLR) Ridge Regression

The developed SMLR model explains about The ridge regression contains all the predictors,
83.76% variation in the yield due to weather whereas other methods consisted of a reduced
parameters. The most critical parameters identified number of predictors thus reducing the model
using SMLR were maximum temperature followed complexity. Ridge regression explains 87.85% of
by morning relative humidity, minimum temperature variation in yield due to all the predictors. The most
and rainfall. This model is considered excellent influential parameter was found to be maximum
according to nRMSE value (Jamieson et al., 1991) temperature followed by rainfall. The developed
ridge model is considered good in accordance with
nRMSE value.

107 | 10-12 | 4
LASSO (Least Absolute Shrinkage and Selection of the variation in yield. The nRMSE value depicted
Operator) that the model performance was excellent.
In LASSO, feature selection is made along
with regularization of parameters, thus preventing
the model from overfitting. The developed model
explains about 87.46% of the variation in yield.
The most influential parameter was found to
be maximum temperature followed by rainfall.
The developed model is considered excellent in
agreement with nRMSE value.
Elastic Net Regression Figure.2. Pearson correlation coefficient between
In ENet method, which is a combination of both groundnut yield and weather indices of Coimbatore
ridge and LASSO, the maximum temperature was district.
found to be the most critical parameter followed by For comparing the performance of SMLR and
rainfall. The developed model explains about 87.48% penalized regression techniques, we used goodness-

Regularization path for ridge regression Cross validation of lambda values - ridge

Regularization path for LASSO Cross validation of lambda values - LASSO

Regularization path for ENet Cross validation of lambda values - ENet

Figure.3. Regularization paths and cross validation of lambda values for ridge, LASSO and ENet
107 | 10-12 | 5
of-fit measures i.e., R2adj, RMSE, MAE, MAPE. The When considering the R2adj, ridge regression was
adjusted coefficient of determination (R2adj) was found to have a better fit compared to other models.
significant for all the models included in the study. The RMSE value of 114.40 was found the least for

Table.4 . Goodness of fit measures obtained for Groundnut yield prediction of Coimbatore district using
different statistical models.
MODEL Adj R2 RMSEC RMSEV MAE MAPE nRMSEV F statistic
MLR 0.8215 117.78 184.09 167.17 0.08 9.64 10.62
SMLR 0.8376 (2) 114.34 178.01 (1) 158.99 (3) 0.08 (1) 9.32 14.18
Ridge 0.8785 (1) 136.51 210.08 (4) 166.93 (4) 1.84 (4) 11.00 5.67
LASSO 0.8746 (4) 136.01 188.09 (3) 158.54 (2) 1.76 (3) 9.85 10.39
ELNET 0.8748 (3) 138.63 181.99 (2) 149.61 (1) 1.65 (2) 9.53 7.77
**Values in parenthesis refers to the rank of the measures.

SMLR followed by MLR and other models. ENet COMPETING INTERESTS

method was found to have a minimum MAE of
There were no conflict of interest in the publication
149.62 followed by LASSO and SMLR. In the view
of this content
of MAPE, SMLR is found to have the least value of
0.082 followed by MLR and other models. However, REFERENCES
the best prediction is not necessarily provided by a
Agrawal, R., Jain, R. C., Jha, M. P., and Singh, D.
model that fits the data well. Hence, we evaluated 1980. Forecasting of wheat yield using climatic
the predictive performance of the models using a variables. Ind. J. Agric. Sci., 50(9):680-4.
validation data set and found that the RMSE value Agrawal, R., Jain, R. C., and Jha, M. P. 1986.
of 178.06 was found to be the least for SMLR, Models for studying rice crop-weather
followed by ENet and other models. In accordance relationship. Mausam., 37(1): 67-70.
with the nRMSE value, all models except ridge were Balabin, R. M., Lomakina, E. I., and Safieva, R. Z.
considered excellent (<10%), whereas ridge was 2011. Neural network (ANN) approach to biodiesel
considered good (10-20%). analysis: analysis of biodiesel density, kinematic
viscosity, methanol and water contents using
CONCLUSION near infrared (NIR) spectroscopy. Fuel., 90(5):
In the present investigation, five different 2007-2015.
multivariate models were compared to predict the Bocca, F. F., and Rodrigues, L. H. A. 2016. The effect of
groundnut yield of Coimbatore district during Kharif tuning, feature engineering, and feature selection
season. The classical Multiple Linear Regression in data mining applied to rainfed sugarcane
(MLR) model shows the presence of multicollinearity. yield modelling. Computers and electronics in
agriculture., 128: 67-76.
So, the performance of other models such as SMLR,
ridge, LASSO and ENet were ranked based on the Cho, S., Kim, H., Oh, S., Kim, K., and Park, T. 2009.
Elastic-net regularization approaches for genome-
different goodness-of-fit measures (Table.4). SMLR
wide association studies of rheumatoid arthritis.
and ENet provide better data fitting, revealing that
In BMC proceedings .3(7) 1-6.
these models can be used for groundnut yield
BioMed Central.,Das, B., Sahoo, R. N., Pargal, S.,
forecasting for the studied region.
Krishna, G., Verma, R., Chinnusamy, V., ... and
ACKNOWLEDGMENT Gupta, V. K. 2017. Comparison of dif ferent
uni-and multi-variate techniques for monitoring
The Agro Climate Research Centre, TNAU and leaf water status as an indicator of water-deficit
Department of Economics and Statistics, Tamil Nadu stress in wheat through spectroscopy. Biosystems
are duly acknowledged for providing the required Engineering., 160: 69-83.
weather data and groundnut yield data. Dhekale, B. S., Sawant, P., and Upadhye, T. 2014.
Weather based pre-harvest forecasting of rice at
ETHICS STATEMENT Kolhapur (Maharashtra). Trends Biosci., 7: 39-41.
No specific permits were required for the Dormann, C. F., Elith, J., Bacher, S., Buchmann,
described field studies because no human or animal C., Carl, G., Carré, G., ...and Lautenbach, S.
subjects were involved in this research. 2013.Collinearity: a review of methods to deal
with it and a simulation study evaluating their
CONSENT FOR PUBLICATION performance. Ecography., 36(1): 27-46.
All the authors agreed to publish the content. Eyduran, E., A. Waheed, M.M. Tariq, F. Iqbal, and S.
Ahmed 2013. Prediction of live body weight from
morphological characteristics of commercial goat

107 | 10-12 | 6
 in Pakistan using factor and principal component Sangun, L., Cankaya, S., Kayaalp, G. T., and Akar, M.
scores in multiple linear regression. J. Anim. Plan. 2009. Use of factor analysis scores in multiple
Sci., 23(6): 1532-1540. regression model for estimation of body weight from
Friedman, J., Hastie, T., and Tibshirani, R. 2009. glmnet: some body measurements in Lizardfish. Journal of
Lasso and elastic-net regularized generalized linear Animal and Veterinary Advances., 8(1): 47-50.
models. R package version., 1(4): 1-24. Sheather, S. (2009). A modern approach to regression
Hastie, T., and Qian, J.2014. Glmnet vignette.,9: 1-30. with R. Springer Science and Business Media.
Hoerl, A. E., and Kennard, R. W. 1970. Ridge Shi, W., Tao, F., and Zhang, Z. 2013. A review
regression: Biased estimation for nonorthogonal on statistical models for identifying climate
problems. Technometrics., 12(1): 55-67. contributions to crop yields. Journal of Geographical
Sciences., 23(3): 567-576.
Iqbal, F., Ali, M., Huma, Z. E., and Raziq, A.2019.
Predicting live body weight of harnai sheep through Singh, R. S., Patel, C., Yadav, M. K., and Singh, K.
penalized regression models. J. Anim. Plant K. 2014. Yield forecasting of rice and wheat
Sci., 29(6): 1541-1548. crops for eastern Uttar Pradesh. Journal of
Agrometeorology., 16(2): 199.
Jamieson, P. D., Porter, J. R., and Wilson, D. R. 1991. A
test of the computer simulation model ARCWHEAT1 Smith, G. 2018. Step away from stepwise. Journal of
on wheat crops grown in New Zealand. Field crops Big Data., 5(1): 1-12.
research., 27(4): 337-350. Tibshirani, R. 1996. Regression shrinkage and selection
Kumar, H. V., Shivamur thy, M., and Lunagaria, M. via the lasso. Journal of the Royal Statistical
M. 2017. Impact of rainfall variability and trend Society: Series B (Methodological.), 58(1): 267-
on rice yield in coastal Karnataka. Journal of 288.
Agrometeorolog.y, 19(3): 286-287. Trnka, M., Hlavinka, P., Semerádová, D., Dubrovsky, M.,
Kumar, N., Pisal, R. R., Shukla, S. P., and Pandey, K. K. Zalud, Z., and Mozny, M. 2007. Agricultural drought
2014. Crop yield forecasting of paddy, sugarcane and spring barley yields in the Czech Republic. Plant
and wheat through linear regression technique for Soil and Environment., 53(7): 306.
south Gujarat. Mausam., 65(3): 361-364. Verma, U., Piepho, H. P., Goyal, A., Ogutu, J. O., and
Kutner, M., Nachtsheim, C., and Neter, J. 2004. Applied Kalubarme, M. H.2016. Role of climatic variables
Linear Regression Models. Richard D. Irwin. Inc.: and crop condition term for mustard yield prediction
Homewood, Illinois. in Haryana. Int. J .Agric. Stat. Sci., 12: 45-51.
Lobell, D. B., and Burke, M. B. 2010. On the Wu, Z., Huang, N. E., Long, S. R., and Peng, C. K. 2007.
use of statistical models to predict crop yield On the trend, detrending, and variability of nonlinear
responses to climate change. Agricultural and and nonstationary time series. Proceedings of the
forest meteorology., 150(11): 1443-1452. National Academy of Sciences., 104(38): 14889-
14894.
Piaskowski, J. L., Brown, D., and Campbell, K. G.
2016. Near infrared calibration of soluble stem Yakubu, A. 2010. Fixing multicollinearity instability in the
carbohydrates for predicting drought tolerance in prediction of body weight from morphometric traits
spring wheat. Agronomy Journal., 108(1): 285- of White Fulani cows. Journal of Central European
293. Agriculture.
Rai, Y. K., Ale, B. B., and Alam, J. 2011. Impact Zou, H., and Hastie, T.2005. Regularization and
assessment of climate change on paddy yield: a variable selection via the elastic net. Journal of
case study of Nepal Agriculture Research Council the royal statistical society: series B (statistical
(NARC), Tarahara, Nepal. Journal of the Institute of methodology)., 67(2): 301-320.
Engineering., 8(3): 147-167.

107 | 10-12 | 7