Browse Prior Art Database

TLOG Monitor Using Real-time Piecewise Linear Regression

IP.com Disclosure Number: IPCOM000030457D
Original Publication Date: 2004-Aug-13
Included in the Prior Art Database: 2004-Aug-13
Document File: 2 page(s) / 47K

Publishing Venue

IBM

Abstract

A real-time least squares model is built from point of sale transaction data. There is no latency between receipt of transaction summary data and update of the model.

This text was extracted from a PDF file.
At least one non-text object (such as an image or picture) has been suppressed.
This is the abbreviated version, containing approximately 52% of the total text.

Page 1 of 2

TLOG Monitor Using Real-time Piecewise Linear Regression

Historically, in the retail industry, transactions logs (TLOG) were sent once daily as a batch from each store to the enterprise. Once the TLOG information was stored at the enterprise, it could be 'data-mined' for store information, such as the store performance. Although this information is valuable, significant information latency is inherent in the process. With the introduction of solutions providing real-time TLOG information to the enterprise, the opportunity is now available to begin to obtain a real-time, rather than just historical, view of a store's performance at the enterprise level.

The IBM WebSphere Business Integration Server - TLOG Processor is the foundation solution for enabling Retail Transaction Summary Log message parsing, processing and writing in the WebSphere Business Integration Message Broker (WBIMB) multiplatform environment.

One of the requirements of the TLOG Processor is to monitor the activity of the middleware, as well as the sales and transaction volume for each store, and to determine whether these levels of performance are statistically acceptable. Questions such as "What is the total volume by store since the store opening? Is that above or below expectation? At any time during the day it should be possible to say that a store should have x number of transactions, plus or minus an expected statistical window, are arising in the field. This is the problem addressed by the TLOG Monitoring system. There is currently no known real-time solution to this problem at the enterprise level.

The proposed system constructs a piecewise linear regression model of POSLog data with minimal redundancy of data capture. That is to say, a least squares estimate is built from real-time data and only relevant accumulated totals are stored rather than duplicating and storing entire POSLog records. Because the model is determined statistically, rather than simply "connecting the dots" each time a sale is recorded, it is a more faithful representation of the daily store revenue and more resilient to sales outliers. In addition, it is clear that a single linear regression for the entire day is inadequate, since the variability of sales would likely be dramatic (and therefore nonlinear) over the course of the day, however, by segmenting the day into one hour windows and calculating a regression over each of these time slots, a reasonable...