A Neural Network Model For Sales Analysis

Published Date: 02 Nov 2017

Predicting Consumer Retail Sales Using Neural Networks

Introduction

Electronic business is based on the sending the information with the help of Internet or Internet applications .As a business electronic shopping is increasing daily .e commerce is total web base application. Mining of e commerce is to provide platform to every business over Internet. The ecommerce application provides selling of services and goods by using network applications such as Internet where we can take the advantage of Internet as an unlimited virtual space for represent goods. Customer has lots of information about the product. But physical verification is not possible, In the field of e commerce data analysis is important .for the required data processing artificial intelligence and neural network is used.

When classification is required for data processing Neural Network can be used in many areas .suppose we consider the data it is not difficult to see that he has to sift out useful information from raw data. Keeping in mind the demands of this domain it is difficult to draw anything related from the available data because of large amount of data.

With the help of neural network, many relations among data set can be find out

Which are unseen previously. In this way with the help of neural network we can find out information which is relevant can be find out very easily. In other word finding information from information is easy by using neural network capabilities the power of MDSS will be greatly enhanced if neural networks can predict sales performance from previous sales and marketing data. [8].

Electronic markets are growing so rapidly that companies must now strive to not only have a presence on the Internet but to differentiate themselves from their competitors.

To attract customers Commerce companies are making personal data collection about customers. and provide the information regarding the products of their interest . We want to describe the role of user profiling within a E commerce website for this we have to use Artificial Neural Network. (ANN) to

Artificial Neural Networks

Aspects of biological neuron system as a gross level are designed with the help of ANN programs. as in human brain individual neurons are connected together through a nerve called axon, similarly ANN are made-up of many processing elements which are tied together with different connections which are differently waited ANN are trained with large data bases and training examples are generated from considerable areas like business data mining and financial predictions it consist of two or more layers of neurons While a predicted output might be whether or not it is going to be sunny. The middle slab contains the "hidden nodes" that provide internal representation for the development of ANN models. For complex models it is often necessary to have a large number of hidden nodes. [5]

Technical issues

A neural network is a parallel system, capable of resolving paradigms that linear computing cannot. ANNs, like people, learn by example. An ANN is configured for a specific application, such as pattern recognition or data classification, through a learning process. Learning in biological systems involves adjustments to the synaptic connections that exist between the neurons. This is true of ANNs as well.

Artificial neural network

An artificial neural network consists of a number of artificial neurons, which are the elementary processing units, which are connected together. Brain is capable of learning by pattern discovery from different set of examples this can be achieved in ANN by adjusting the weights or the synaptic interconnection strength between neurons, on a given predefined learning algorithm. It is very important to see while building a neural network what are the inputs how many layers we are using and the activation function to be implement [3]

The ANN is identified by it processing algorithm and its architecture and learning algorithm processing algorithm and its learning algorithm .A neural network model for sales analysis neurons and the way they are connected. The processing algorithm specifies how the neural network with a given set of weights computes the outputs for any inputs set.

The learning algorithm gives the idea about how weights are adapted by the network based on the training sets .the trained network is capable to classify the input sets in to resembling category **in the way that they are presenting same distinctive features. This generalization capability, improved with the ability to deal with imperfect or incomplete data, is highly useful in real world applications.

Data miss match may happen many times that mines the input data is not always fixed in pattern and the decision must be taken depending on present data ANN has the ability to learn from examples and take advantage of suitable relationships between set of data .this is very useful when data is in huge amount in other words, there are many potential parameters that can affect the final solution. Besides that, artificial neural network’s adaptation makes possible on the run learning, which will avoid the premature ageing of intelligent system’s knowledge. [4]

Marketing strategies

In business marketing is a very important domain that informed about the value and target of achievement of company and its products, it also gives provides feedback of customers about products ‘‘Value" is worth derived by the customer from owning and using the product. Marketing informs about task related information of competitors also. And there is no such a strategies about marketing, how to collect the information and compare it with present information, and what information is need to deliver and to whom and how much? When to deliver and where to deliver. Once the decisions are made, there are numerous ways (tactics) and processes that could be employed in support of the selected strategies. [4]

The main focus of marketing is to develop and maintain a plain for a company and the products of it within the target area of market. We must develop a plan for a company which is sustainable and profitable to maintain relation with customers. To achieve this goal all business domains are same responsible but marketing caries larger part of it.

In business customer always an important role in any business we have to provide offers to customer additional options Up-selling is a sales technique whereby a salesman attempts to have the customer purchase more expensive items, upgrades, or other add-ons in an attempt to make a more profitable sale. Upselling can imply selling something additional, or selling something that is more profitable.

The neural network model is used in marketing for making decisions and gives prior information regarding the quantity and sale regarding the revenue. Data require for this are collected by the marketing and sale performance of that particular product in terms of revenue we can maintain relationship between marketing variables and revenue ,one hidden layer networks were used following the authors’ experiences. This can be considered as a rule of thumb. The number of nodes used in hidden layers is around half of the number of input variables of a neural network. [6]

Neural Networks For Consumer Retail sales Prediction

Retail sales is one of the most important domain in business and if we are able to give some ideas in advance related it will be more beneficial for strategic and planning decisions in effective operations of retail businesses and retail supply. The effectiveness of data pre-processing such as forecasting of sales and revenue generation .we can say that without pre-processing ANN are not able to give any result with related to sale and trade. [9]

For new business trades neural network provides solutions and used as a powerful tool. Neural Network software gives different solutions to wide range of business applications across many fields. The business man can easily use these tool of Neural Network in to their on line applications to check the online competition and different offers given by the compotators for which customer is looking for

For gating ideas about price, margins on price, market trades, inventory etc. N.N. applications are useful .these applications are also helpful to give prior information about market fluctuations and risk .We are able to design stargazes depending on these results by time. In our example we are designing such N.N. which helps us to provide productive business.

Neural network applications that define manufacturing processes for best chances of success, modeling the physical environments that productive businesses are always trying to optimize. for interactivity.

Managing Daily and Weekly data with Back Propagation Rules

We have to cover various product categories for marketing point of vive and promotion .And calculate the expenses for a desired time period recurred for the same. Different media, expenditure levels and sales data are collected for the period .This will be defined as a marketing expenditure. This set of data is acting as an important link between advertising expenditure and effect of it on sales and profit and improvement of the marketing department of the origination.

In retailer organization we are studying 4 different sections in advertising and promotion. the sales promotion and special event section plays an important role while two other plays as supporting .we can take decision from past results of promotion and campaigns then combined with budgets, inventory data., figures of sales and profit .Depending on this budget is adjusted from the previous year figures .From the data generated we want to build a neural system which helps us to predict net sale values .companies production dept. and sailing dept. will gate some additional future knowledge from these results .this is why important for management point of view .To obtain revenues we have to study the ratio of sales and advertising cost .To get best results from N.N. we have to develop it for following input values

Net sales values in last week

Monthly average rate of product

Monthly average rate

Impact of promotion on sales.

In back propagation method the configuration of N.N. is important issues

There are many observations showing that back propagation N.N. is an efficient method used for function approximation. Which are given to a number of hidden nodes with sigmoid output but these we can’t specify the no of hidden units needed.

With regard to the number of units and Weights, that the number of weights for a satisfactory network performance should be less than one tenth of the number of training patterns. An obvious shortcoming of this rule is that it is very restrictive for limited data sets and large Networks. A cross-validation procedure may be needed due to the large number of parameters provided for an oversized network basically we want to carry out some different calculations to find out the N.N. for particular problem. For number of hidden layer thumb rule is used which defines the no of hidden units as ½ of the sum of the i/p and o/p units .there for we are using in this study for the daily model. There are no defined rules of thumb for a network with 2 hidden layers with respect to the number of units in first and second layer.

Preprocessing the Input Data

We have to process the data efficiently to input it in the neural network .The good idea is to transform the raw time data in to specific indicators that represents the information more clearly. The sales information must be scaled which supports learning algorithm (Back propagation algorithm) the sales information must be scaled to The scaling is necessary to support the back-propagation learning algorithm We tested several scalings for the sale and the turnover time

in the above example we take the information of price and marketing cost and compare them .from the records of supermarket . we use the time series of the turnover of all present items which comes in same product category this calculation is for the term of 31 months

In this example we are using feed-forward MLP network with one hidden layer with back propagation training method for prediction of sales we are using the last month’s data as input for I/P layer form this we want to predict the sale in next month[12]

[3]

In feed forward perceptron if we provide suitable amount of training it can approximate any function so known values of this function are presented to the network .It is expected that the net should learn the function rules .when we change the basic values and the weights, behavior of the net must be changed .

When we want to in put the data in to the net required processing of data is must. The data must be normalized in 0, 1 before it act as a input. In our example assume that the information is for T (time) of previous weeks

[3]

For each article i and recent week t we use a three-dimensional vector:

For a week t in the future the vector is reduced by the unknown sale

To predict the sale for one article within a week t. we use a window of the last n weeks. So we have the following input vector for each article i:

All the products are belongs to the same group .when sale of one product (article) is increased it effects on the sale of other product. Because of this we summaries the I/P vector of all p articles to get the vector given to input layer

The sale of article (i) within week (t) Is the nominal value in the output layer[3]