Surety is performing system maintenance this weekend. Electronic date stamps on new Prior Art Database disclosures may be delayed.
Browse Prior Art Database

A new approach to improve the accuracy of merchandising forecast based on Naive Bayes analysis

IP.com Disclosure Number: IPCOM000235963D
Publication Date: 2014-Apr-01

Publishing Venue

The IP.com Prior Art Database


For single promotional product, find out the independent and related variables, using Naive Bayesian classification for data mining directly, provide instruction to the following up business plan decision based on the given data. Upgrading Naive Bayesian classification algorithm logic, introduce the deviation as the naive Bayes classifier weights, make it more comprehensive support for the actual promotional activities.

This text was extracted from a PDF file.
This is the abbreviated version, containing approximately 39% of the total text.

Page 01 of 10

A new approach to improve the accuracy of merchandising forecast based on Naive Bayes analysis


Currently retail merchandising forecast is using different variables with different weight to set up business model to do optimization. It had been proved that during the setup of entire business model, the most important factor is the model creator's industry experience. Meanwhile the optimization accuracy will be different with using different variables and weight.

This article will introduce to use of the historical optimization result data, establish promotion or pricing models from data mining way instead of modeling. We plan to use Naive Bayesian classification forecasting techniques to do analysis of the historical optimization result data, other than from the perspective of modeling, in another aspect to explore effective information for modeling strategy, provide more effective reference information and scientific basis for decision making, at last to increase the modeling accuracy.

Naive Bayesian classifier prediction work processes

Each data item is an n-dimensional attribute value X {X1, X2... Xn} indicates that properties of the n A1, A2... An.

Category tag attributes C = {y1, y2... yn}.

Calculating P (y1 | X), P (y2 | X)... P (yn | X). The Calculating logic as below:

In naive words it can be expressed as:

denominator can be negligible)




(The denominator does not depend on the value of a given characteristic, so the


Page 02 of 10

We can retrieve all the optimization result data from database, with deep data mining, these data can provide potential and useful information to optimize the new promotion modeling. Data collection should collect marketing data as much as possible, such as promotion price floating rate, promotion time of duration, promotion type, and whether it is the seasonal promotion, whether it is the major festivals promotion. According to the collected promotional materials, we can define which one of the promotions is the good quality promotion and which one is bad. Firstly, we do cluster analysis to screening out noise information.

Cluster analysis refers to the physical or abstract objects, grouped similar objects into several classes. The goal is to collect data based on similarity. By cluster analysis, some obvious noise information will be filtered, it will avoid too large or too far away individual characters, ensure the accuracy of the data collection.

Product based prediction


Page 03 of 10


Page 04 of 10

Here take result data from one large retailer (A) for example, these promotional data are based on our product with using modeling. We plan to use Naive Bayesian prediction to analysis the historical promotion result data, generate reference to adjust the next promotion modeling strategy.

The analysis as below:

Data item is X{X1, X2... C}, these are promotion price floating rate, promotion time of duration, promotion type, promotion efforts and C for promotion category tags....