See discussions, stats, and author profiles for this publication at: [Link]

net/publication/258652104

Sales Forecast of an Automobile Industry

Article in International Journal of Computer Applications · September 2012

DOI: 10.5120/8474-2403

CITATIONS READS

24 8,734

2 authors:

Dr. Rashmi Sharma Ashok Sinha

Ajay Kumar Garg Engineering College ABES Engineering College
24 PUBLICATIONS 52 CITATIONS 13 PUBLICATIONS 313 CITATIONS

SEE PROFILE SEE PROFILE

All content following this page was uploaded by Dr. Rashmi Sharma on 15 October 2020.

The user has requested enhancement of the downloaded file.

 International Journal of Computer Applications (0975 – 8887)
Volume 53– No.12, September 2012

Sales Forecast of an Automobile Industry

Rashmi Sharma Ashok K. Sinha
ABES Engineering College Ghaziabad ABES Engineering College Ghaziabad

ABSTRACT research and development; quality of service, pricing and

Sales forecast plays a prominent role in business strategy for financing policies; and public image. Forecasters also evaluate
generating revenue. Sales forecast depends on some of the the quality and quantity of the customer base to determine
factors as the market demand, promotion strategy used, living brand loyalty, response to promotional efforts, economic
standard of the people, inflation rate, consumables price, viability, and credit worthiness. The condition of the overall
public image of the company, market share, quality of service economy is still a primary determinant of general sales
and so on. In this paper sales forecast of Maruti Suzuki Ltd, volume. If the prices for products have changed over the
an automobile industry in India is considered. The inflation years, changes in volume of sales may not correlate well with
rate, petrol price, previous month sale are found to be more volume of units. At one point in time when demand is strong,
prominent parameters influencing the sales forecast of cars in a company raises its prices. At another time, a company may
this company. The model is trained using Fuzzy Neural Back engage in discounting to draw down inventories. This process
Propagation Algorithm. The result thus obtained is compared is similar to an inflation index, which provides constant
with other statistical technique like multiple regression prices. Forecasters study the underlying assumptions of trend
technique. However the result obtained by proposed algorithm variations to understand the important relationships in
is found to be superior to the result obtained by multiple linear determining the volume of sales. By analyzing month-to-
regression technique. month trends and seasonal variations over both the long and
short terms, business owners and managers can adjust the
sales forecast to anticipate variations that historically repeat
General Terms themselves during budget periods. Forecasters also trend
Neural Networks, Fuzzy System, Decision Support System, individual products, using indexes to adjust for seasonal
Multiple Linear Regressions fluctuations and price changes. Product trends are important
for understanding the life cycle of a product. Consumer
Keywords attitude and lifestyles anticipate product introductions and
Sales Forecast, FBPN, Non-linear method, Automobile
technological changes. Demand based on anticipation is
Industry
becoming the dominant feature of the technological age. The
rapid pace of technological development and new product
1. INTRODUCTION introductions have shortened product life-cycles.
The automotive sector is one of the core industries of the
Indian economy. After allowing continuous economic The most commonly used techniques for sales forecasting
liberalization in since 1991 the auto industry has witnessed a include statistically based techniques like time series,
phenomenal growth in the last two decades. India has regression techniques and computational intelligence method
attracted many global automotive players in recent years. The like fuzzy systems, artificial neural networks and neuro-fuzzy
industry has greatly benefitted from increase in the paying systems [9,10,11,12]. Several forecasting techniques have
capacity of the consumers; this has led to an increase market been developed, each one with its particular advantages and
demand. disadvantages compared to other approaches.

For the sales forecasting to be accurate, managers need to Pei-Chann Chang, Yen-Wen Wang [1] proposed sales
consider all or some of the following factors: historical forecasting in PCB industry using Fuzzy Delphi and back-
perspective, business competence, market position, generic propagation model and demonstrated effectiveness of the
economic conditions, price index, intra-company trends, fuzzy back propagation network (FBPN) that is superior to
product trends, sources and magnitude of product demand. other traditional approaches. The input data is divided into
three domains- Market demand domain, Macroeconomics
A sales forecast predicts the value of sales over a period of domain and Industrial production domain.
time. It becomes the basis of marketing mix and sales
planning. Sales forecasting is crucial because without a proper Mirnaser Mirbagheri [2] recognized the effective variables
sales forecast a company cannot program to attain the desired which effect economic growth in Iran, and then applied
sales and marketing objectives. It is based on a number of appropriate means for modeling and forecasting the main
assumptions regarding customer and competitor behavior as macroeconomic variables.
well as the market environment, and therefore, its reliability Stefka Stoeva, Alexander Nikov [5] presented an extension of
depends upon a number of uncertain parameters. the standard backpropagation algorithm (SBP).In this research
Management analyzes previous sales experience by product SBP and FBP are compared .Fuzzy back propagation
lines, territories, classes of customers, and other relevant algorithm shows considerably greater convergence rate
details. Management needs to consider a time line long compared to SBP algorithm without oscillations.
enough to detect trends and patterns in the growth and the Gholamreza Jandaghi, Reza Tehrani, Davoud Hosseinpour,
decline of sales volume. The ability of a company to respond Rahmatollah Gholipour and Seyed Amir Shahidi Shadkam
to the results of a sales forecast depends on its production [6] proposed a model based on Fuzzy-neural networks for the
capacity, marketing methods, financing, and leadership, and prediction of stock price. The prediction was done by the two
its ability to change each of these to maximize its profit linear and nonlinear models for one ahead and multi ahead in
potential. Sales forecasting also considers the competitive stock price by using exogenous variable of stock market cash
position of the company with respect to its market share; index, and the results show the preference of nonlinear neural-

25
 International Journal of Computer Applications (0975 – 8887)
Volume 53– No.12, September 2012

Fuzzy model to classic linear model and verify the capabilities

of Fuzzy-neural networks in this prediction.
Bias
Yan Liu, Min Sun [7] proposed the use of fuzzy optimization
Sales Sales Forecasted
BP neural network as a management tool for the maintenance Petrol Rate Forecast
of expressway pavement.
Inflation Rate System

2. METHODOLOGY
1. The significant parameters of the model are identified
2. The proposed sales forecast model is represented by
artificial neural network (ANN).
3. The ANN model is trained by past data on inputs and the Sales as input for next
outputs. month
4. The model error is compared with the error obtained by Figure 1: Sale Forecast System
multiple regression technique.
5. Using the converged values of interlayer coefficients 2.2 Neural Network Architecture
sales forecast can be computed. The neural network model used in the study is the multilayer
perceptron (MLP) also known as a supervised network. This
2.1 Parameters of the Model network requires a desired output in order to learn, achieved
It is important to examine some of the significant factors that by creating a model that correctly maps the input to the output
affect the sale of goods. using historical data so that the model can then be used to
1. Price of the Product- There is inverse relationship produce the output when the desired output is unknown. This
between price and the amount consumers are willing and uses three-layer architecture: input (known parameters),
able to buy is often referred to as the law of demansd. hidden layer, and output layer (known value)
2. The Consumer's Income -For most goods, there is a
positive (direct) relationship between a consumer's
income and the amount of the goods that one is willing
and able to buy. Input Hidden Output
3. Consumer Awareness- The percentage of population or Layer Layer Layer
Bias
target market, who are aware of the existence of a given
brand or company.
4. Inflation rate - The fall in the value of local currency in Petrol rate
the global market. Sales Forecasted
5. Petrol/ Diesel Rate - The increasing rate of the crude oil
in international market affects the petrol / diesel rates in Inflation rate
India.

In the context of Maruti Suzuki India

1. Price of the product is not considered, as the company Sale of previous month
has good market value.
2. These days ample loan facilities are available so the
consumer’s income also does not directly affect the sale
of the product. Figure 2: Neuro fuzzy Model for Sale Forecast System
3. Consumer’s awareness is evident from the past sales of
the product. 2.3 Model implementation in Neural
4. Inflation rate affects the sale as it directly affects the loan
installments and the resale value of the product.
Network
A sample of data on Inflation rate, Petrol rate, actual sales of
5. Diesel/Petrol is a consumable in the product and required
previous month are shown in the table I.
on daily basis. So it affects monthly expenditure of the
consumer.
Neural network parameters are:

The significantly affecting parameters considered in our model Number of patterns for training - 11
are: Nodes in input layer - 4 (Three parameters and one bias)
1. Inflation Rate Hidden layer - 1
2. Petrol/Diesel Price Nodes in hidden layer - 2
3. Sale of the previous month. Nodes in output layer - 1
Weights between input layers and hidden layers - 8
Weights between hidden layers and output layer - 2
Learning rate (η) - 0.3
Constant value (α) - 0.5
Number of iterations - 50
Software used – Program (m file) developed on MATLAB 7.0

26
 International Journal of Computer Applications (0975 – 8887)
Volume 53– No.12, September 2012

Table 1. Actual data considered for the system training 2.4 Model implementation in Multiple
( Sales data For Year 2011)
Linear Regressions
Curr The directly affecting parameters are the same:
Inflati Previo ent 1. Inflation Rate
S. Petrol on us mont 2. Petrol/Diesel Price
No. Month Price Rate month h 3. Sale of the previous month.

1 January 58.37 9.47 66290 74355 The equations used are for 100% confidence
2 Februray 58.37 9.3 74355 74802 [b, bint, r, rint, stats] = regress(y, X);
For 100% confidence level the residuals lies between +0.35 to
3 March 58.37 8.82 74802 81375 -0.35
4 April 58.37 8.82 81375 59971 The equations used are for 50% confidence
[b, bint, r, rint, stats] = regress(y, X, 0.5);
5 May 63.37 9.41 59971 65237 For 50% confidence level the residuals lies between +0.2 to -
6 June 63.37 8.72 65237 54422 0.2
Here X is calculated by
7 July 63.7 8.62 54422 47127
8 August 63.7 8.43 47127 53539 X=[ones(size(x1)) x1 x2 x3];
Where X1 is Inflation rate,
9 September 66.84 8.99 53539 57049 X2 is Petrol rate
10 October 66.84 10.06 57049 35868 X3 is Sales of previous month

11 November 66.42 9.39 35868 61080 Here y is the actual sale of the month
Residual Case Order Plot
Mean Square Error (MSE) is calculated as follows:
0.3
1 n
MSE 
n t 1
( Dt  Ft ) 2  0.2

0.1
Where
Residuals

Dt is predicted by the individual program for a pattern t 0

Ft is the targeted value for a pattern t

-0.1
n is the total number of pattern
-0.2
-3
x 10 Plot of MSE
9 -0.3

8
1 2 3 4 5 6 7 8 9 10 11
Case Number
7

6
Figure 4: Residue case order plot (for 100% confidence
level)
percentage error

4
Residual Case Order Plot

3 0.2

2 0.15

1 0.1

0 0.05
1 2 3 4 5 6 7 8 9 10 11
Residuals

patterns
0

Figure 3: Graph plotting the patterns vs. % error -0.05

-0.1
Mean Square Error (MSE) = 7.7728e-006
-0.15

In the first stage the model is trained with set of known input -0.2
values of sale with random weights assigned. The error 1 2 3 4 5 6 7 8 9 10 11
Case Number
measures (actual value minus computed value) are distributed
to the elements in the hidden layers using back propagation.
Different weights connecting different elements in the Figure 5: Residue case order plot (for 50% confidence
network are corrected till the values converge within level)
acceptable range.

27
 International Journal of Computer Applications (0975 – 8887)
Volume 53– No.12, September 2012

Table 2. Comparison of Minimum and Maximum values IEEE’97 page no.1381-1386Tavel, P. 2007 Modeling
of both models and Simulation Design. AK Peters Ltd.
[4] Jingtao Yao, Nicholas Teng, Hean-Lee Poh , Chew Lim
Tan(1998) Forecasting and Analysis of Marketing Data
Maximum Minimu Using Neural Networks Journal Of Information Science
Range
Value m value And Engineering 14, 843-862.
[5] Stoeva Stefka, Nikov Alexander(2000) .A fuzzy
Neural Network -3 -3 backpropagation algorithm. Fuzzy Sets and Systems 112
8.3X10 0 8.3X10 (2000) 27-39.
Model
100% [6] Gholamreza Jandaghi, Reza Tehrani, Davoud
Multiple Hosseinpour, Rahmatollah Gholipour and Seyed Amir
confidenc 0.3611 -0.3626 0.7237
Linear Shahidi Shadkam ,Application of Fuzzy-neural networks
e
Regressi in multi-ahead forecast of stock price African Journal of
50%
on Business Management Vol. 4(6), pp. 903-914, June 2010
Confidenc 0.2079 -0.1956 0.4035
Model
e [7] Yan Liu, Min Sun Fuzzy Optimization BP Neural
Network Model for Pavement Performance Assessment
Proceedings of 2007 IEEE International Conference on
3. CONCLUSION Grey Systems and Intelligent Services, November 18-20,
The present work has successfully implemented sales forecast 2007, Nanjing, China.
model of an automobile industry using Fuzzy BPN algorithm.
The result obtained by neural network is further compared [8] Jason E. Kutsurelis, Forecasting Financial Markets Using
with the result obtained by multiple linear regression analysis. Neural Networks: An analysis of methods and accuracy,
The result obtained by multiple linear regressions have range Thesis September 1998 , Naval Postgraduate School
of 0.7237 for 100% confidence and 0.4935 for 50% Monterey, California.
confidence. The result range obtained by proposed algorithm
[9] Rajasekaran S,Vijayalakshmi G.A.,”Neural
is of order 8.3X10-3 and is found to be superior to the result
Networks,Fuzzy Logic and Genetic Algorithms”,PHI,
obtained by multiple linear regressions. The error obtained by
2003.
neuro-fuzzy BP architecture is 7.7728e-006.
[10] Little, John D. C., Decision Support Systems for
4. REFERENCES Marketing Managers, Journal of Marketing (pre-1986);
Summer 1979; 43, 000003; ABI/INFORM Global pg. 9.
[1] Chang Pei-Chann , Wang Yen-Wen (2006). Fuzzy
Delphi and back propagation model for sales forecasting [11] Horris C. bung, (1995) Neural Network in supply chain,
in PCB industry Expert Systems with Applications 30 Engineering Management-Conference, IEEE, pg 347-
(2006) 715–726. 352.
[2] Mirbagheri Mirnaser(2010). Fuzzy-logic and Neural [12] A. Weigend and N. Gerschenfeld, Eds., Time Series
network Fuzzy forecasting of Iran GDP growth. African Prediction: Forecasting the future and Understanding the
Journal of Business Management Vol. 4(6), pp. 925-929. past reading, MA: Addision –Wesley, 1994.
[3] Escoda , [Link], ASanz , [Link] (1997) Demand
Forecasting by neural fuzzy technique IEEE, FUZZ-

28

View publication stats