Browse Prior Art Database

Apparatus to model impact of business events as a combination of modular statistical patterns

IP.com Disclosure Number: IPCOM000236955D
Publication Date: 2014-May-23
Document File: 6 page(s) / 86K

Publishing Venue

The IP.com Prior Art Database

Abstract

Pattern recognition in statistical domain is a challenging problem. A variety of tools and techniques leverage the historical data to the future forecast. These tools and techniques differ in how they decompose a signal into different periodic and aperiodic patterns. Any process within a system is subjected to internal and external events, periodic and aperiodic, and are characterized by time of occurrence, duration and distribution. These events sometimes come in complex patterns/distribution and can have either a transient or permanent impact, and can deviates the behavior of observed process from its natural course e.g. marketing campaigns affecting call volumes to a call center, festivals like Ramzan affecting economy, etc. For systems which are not operating in steady state it is challenge to identify the natural course. In this article we show how to model an event and assess its impact for processes close to steady state. We also explained method and system showing how and where to assign these events in future so that appropriate adjustments to the forecast could be made.

This text was extracted from a PDF file.
This is the abbreviated version, containing approximately 42% of the total text.

Page 01 of 6

Apparatus to model impact of business events as a combination of modular statistical patterns

How to model a business event and assess its impact for processes close to steady state. How and where to assign these events in future so that appropriate adjustments can be made to the forecast. A system and method to identify complex patterns as a function of known modular patterns. A system and method to estimate future event distribution from the combination of modeled events in history

Problem Statement

Pattern recognition in statistical domain is a very challenging problem. A variety of tools and techniques, leverage the historical data i.e.

historical patterns to forecast for future.

    These tools and techniques differ in how they decompose a signal into different periodic and aperiodic patterns.

    Any process within a system is subjected to internal (to a system) and external events

    Events (both periodic and aperiodic) deviates the behavior of any observed process from its natural course

    Event is characterized by its time of occurrence, duration and distribution.

    These events sometimes come in complex patterns/distribution and can have either a transient or permanent impact e.g. Marketing campaigns affecting call volumes to a call center, festivals like Ramzan affecting economy, etc.

    For systems which are not operating in steady state it is more challenging to even identify the natural course as well.

Main Issues:

    How to model an event and assess its impact for processes close to steady state

    How and where to assign these events in future so that appropriate adjustments can be made to the forecast.

    A system and method to identify complex patterns as a function of known modular patterns. Following methods and systems are used:

    Method 1 : Identify outliers (known modular patterns) in the historical data based on the best fit statistical model. (automated)

    System 1 : Design of a system which allows user to view these modular patters called as outliers in the historical data (automated)

Method 2: Method to assess net impact of events as a convex

1


Page 02 of 6

combination of these individual outliers at any time point identified in Method 1. (automated)

    System 2: Design of a system which allows business user (not even skilled) to model business event (Internal or external) from the combination of outlier impact as suggested by method 2. (Semi-automated)

    A system and method to estimate future event distribution from the combination of modeled events in history

    Method 3: Method to estimate future event of similar nature from the combination of historical events modeled in method 2.

    System 3: Design of a system to incorporate events (deterministic) for future forecast as suggested by method3. (Semi-automated)

Methods

    Method 1: Design of a system and method to identify events (with complex patterns) as a function of known modular patterns.

    Method 1 : Identify outliers (known modular patterns) in the historical data based on the best fit sta...