Business Processes, Dynamic Contexts, Learning

Business Processes, Dynamic Contexts, Learning

Michael M. Richter (University of Calgary, Canada)
Copyright: © 2014 |Pages: 11
DOI: 10.4018/978-1-4666-5202-6.ch037
OnDemand PDF Download:
$30.00
List Price: $37.50

Abstract

In this article we present relations between complex business processes and machine learning techniques. The processes considered here are mostly related to planning. Planning takes place in preparing many decisions and often it is encountered with a rapidly changing context that constitutes an open world. The underlying structure and preconditions of the processes is quite often not known and hence the processes are regarded as stochastic. One can only observe the processes. Such observations deliver data and these data contain some knowledge about the processes in a hidden form. As a consequence, machine learning methods are involved here. The idea is to give the business persons an overview of quite different machine learning techniques so that they can select suitable ones. We provide a number of examples for business processes that we use for illustrations.
Chapter Preview
Top

Introduction

In this article we relate machine learning methods to business processes. In business are quite many unstructured data that contain hidden information. Business is concerned not only with internal decisions in a closed world. It is rather confronted with a rapidly changing context. That means we deal with an open world.

Dynamic processes and workflows can be changed before or during the execution by a human. This does not allow a precise requirement analysis.

Such changes may happen for different reasons, mainly caused by unpredicted external events, e.g.:

  • External changes

  • New information during the execution

  • The execution shows that the goals cannot be achieved in the intended way.

All these topics give rise to learning because they are not explicitly given but rather hidden in some data. An example is payment behavior analysis. It deals with a certain customer who is partially unknown to us and we want to predict how the customer behaves in the future. However, we have quite a number of data about this customer which contain knowledge in a hidden form. The purpose of learning is to make this knowledge explicit in such a way that we can make decisions about who to treat this customer in the future. Such decisions are based on the insight in customer’s behavior and can be regarded as a classification problem. Such examples are discussed below.

A difficulty is that even the structure of the processes is mostly partially unknown and is regarded as stochastic. The only accessible knowledge is observations about the processes executed so far and about the external world. To formulate them is an essential goal for further planning. This leads to the area of stochastic learning. The reader is assumed to be familiar with general business processes and some general (incomplete) view on machine learning but want to get a comprehensive overview.

In the remaining sections we proceed as follows:

  • Concepts and terminology of general processes and planning.

  • Businesses and machine learning.

  • A structural overview of machine learning techniques and methods.

  • Major examples of business activities and their relations to machine learning.

It is not intended that the reader learns business or machine learning. Therefore we do not provide detailed definitions of concepts and methods. The idea is that business people get an overview of machine learning so that they can choose a method for their application. Then they can go deeper into the method by investigating the literature.

Top

Processes

Processes, Plans, and Actions

The difference between processes and plans is that a process describes what happens and a plan says what could happen. In a sense, processes are realizations of plans. In processes as well as in plans one uses actions for the description. Actions are in their general form mappings:

states → states.

States are described by description elements. In a simple way these are attributes and their values. Actions have preconditions that determine when their executions are allowed. Such allowed actions are arranged in a so-called plan graph with the elements

(state t1, action a, state t2).

A plan p is represented as a totally ordered sub graph of the plan graph, i.e. it is a path in the plan graph:

state1→state2→⋯→staten

A process models describe sets of processes. Formally one defines them as partially ordered sets actions that may contain variables.

A planning problem is a pair initial state and goal state. A solution S to a planning problem is a plan p where state1 is the initial state and staten is the goal state. In many situations in business the problem cannot be stated in such a simple way. The two major reasons for this are:

  • 1.

    Some knowledge is missing.

  • 2.

    The context is changing that requires replanning and interference of planning and execution.

If the needed knowledge is not explicitly available one has no other method than to obtain it by machine learning methods from the knowledge hidden in the data.

Intensive research on business runs under the name process mining. It is performed by an IEEE task force group that produces the “Process Mining Manifesto,” see (Aarts 2011).

Key Terms in this Chapter

Markov Process: A stochastic process where the probabilities of the events depend on the previous event only.

Hidden Markov Models (HMM’s): Markov models with unknown probabilities.

Case-Based Reasoning: Technique to store solutions for solving actual and similar problems.

Context: Contains everything of interest that can influence a decision.

Dynamic Processes: Processes structure and parameters can change at each time.

Dynamic Time Warping: Compares two dynamic time series with respect to their similarities.

Bayesian Reasoning: Stochastic reasoning based on conditional probabilities.

Complete Chapter List

Search this Book:
Reset