�FUZZY COGNITIVE MAPS AND PRODUCT PLANNING THROUGH BUSINESS INTELLIGENCE��

By Nikolaos Zervos and Peter P. Groumpos

University of Patras, Greece.

​

Presented by Prof. Peter P. Groumpos groumpos@ece.upatras.gr

​

���The Hellenic Society for Systemic Studies (HSSS) �15th HSSS National & International Conference�Systemics and Business Intelligence��Department of Informatics �University of Piraeus�29-30 November 2019���� ���

​

Presentation Overview

• Introduction
• Problem Statement
• Business Intelligence (BI)
• Product Planning
• Fuzzy Cognitive Maps (FCM)
• Case Study
• Comments - Conclusions
• Future Extensions

​

​

​

INTRODUCTION

• Production planning is a process used by manufacturing companies to optimize the efficiency of their processes.

​

• Effectively utilize limited resources in the production of goods so as to satisfy customer demands and create a profit for investors.

​

• Resources include the production facilities, labor and materials.

​

• Constraints include the availability of resources, delivery times for the products, and management policies.

​

​

PROBLEM STATEMENT

​

• Why is it important to have a carefully developed production plan?
• Firms need to have a production planning strategy to ensure that there is sufficient capacity to meet the demand forecast and to:
• Minimize changes in production rates
•   Minimize changes in work-force levels
•   Maximize the utilization of plant and equipm

​

CAN THESE BE ACHIEVED

WITH CLASSICAL APPROACHES?

HOW INTELLIGENCE, AI and BI CAN BE USEFUL?

THE NEW FUZZY COGNITIVE MAPS ARE PRESENTED AS A NEW EFFECTIVE AND EFFICIENT APPROACH

​

​

Business Intelligence (1/2)

• Business intelligence (BI) is a technology-driven process for analyzing data and presenting actionable information which helps executives, managers and other corporate end users make informed business decisions.
• The processes, technologies and tools needed to turn data into information and information into knowledge and knowledge into plans that drive profitable business action.
• BI encompasses a wide variety of tools, applications and methodologies that enable organizations to collect data from internal systems and external sources, prepare it for analysis, develop and run queries against that data and create reports, dashboards and data visualizations to make the analytical results available to corporate decision-makers, as well as operational workers.

​

​

Business Intelligence (2/2)

• There are several issues inherent to any BI project:
• Data exists in multiple places
• Data is not formatted to support complex analysis
• Different kinds of workers have different data needs
• What data should be examined and in what detail
• How will users interact with that data

Product Planning (1/3)

​

Many critical decisions are made when designing a production system. One of them relates to the product that the system will produce, the result of the process of transforming raw materials (inputs) into useful outputs. It is a strategic decision, ie it has long-term effects on the system and largely identifies other system parameters.

The purpose of product planning is to produce products that will be treated favourably by customers and have competitive prices.

Product Planning (2/3)

• The process of developing a product is the sequence of steps or actions a company uses to capture, design, supply, or manufacture and commercialize a product. Many of these steps / actions are more theoretical and organizational than natural.
• Each organization uses processes that are not only different from those of other organizations, but often differ (within the same organization) for each of the different types of development programs.

Product Planning (3/3)

This process is complex and time consuming and its main phases are:

1. Search for goals and ideas
2. Selection of ideas
3. Preliminary design of the product
4. Manufacture of prototype
5. Sales forecast
6. Market test
7. Product development
8. Market introduction of the product

Fuzzy Cognitive Maps (1/6)

​

​

Cognitive Maps were proposed by R. Axelrod in 1976 as a way of representing social scientific knowledge as well as modeling decisions in social and political systems. Since then, Cognitive Maps have been applied in many scientific fields.

Fuzzy Cognitive Maps (2/6)

• A Fuzzy Cognitive Map or Fuzzy Cognitive Map (FCM) consists of nodes representing the parameters (inputs, outputs, states) of the system under study and receiving values ​​in space [0,1].
• The nodes are interconnected. These interfaces express the existing cause-effect relationship. The direction of the interface indicates whether the value of node Ci affects the value of node Cj, or vice versa.
• The value of the weight of a Wij interface expresses the degree to which the node Ci affects the value of the node Cj and receives a value in the interval [-1,1].
• The sign of the weight of a Wij interface indicates whether the relationship between nodes is proportional, that is, an increase in the value of one node causes an increase in the value of the other node or not. Specifically:
• Wij> 0 positive causality between the two nodes
• Wij <0 negative causality between the two nodes
• Wij = 0 no relationship between the two nodes

Fuzzy Cognitive Maps (3/6)

​

• The procedure followed for selecting and developing an FCM is as follows:
• Stage I: Selection of the number N and the type of Ci nodes that make up the FCM.
• Stage II: Determining the correlation between nodes, which node affects whom.
• Stage III: Type of correlation between nodes, positive Wij> 0, negative Wij <0 or no Wij = 0.
• Stage IV: Determine the degree of correlation between two nodes, what is the value of the Wij interface weight between the nodes.
• Stage V: Combining Expert Opinions and Final FCM Planning

Fuzzy Cognitive Maps (4/6)

FCMs are a promising modeling methodology, especially for highly complex systems that are non-linear and contain ambiguous situations. But with classical mathematical models (Type I, II, III) some drawbacks emerge. These disadvantages create the need to use a new approach to FCMs, while keeping the core of the method intact.

​

Key Disadvantages - Limitations of classic DRC models are:

1. Lack of knowledge of the system
2. Expert dependency
3. Inability to self-educate
4. Unclear causality
5. Calculation equation
6. Use of Sigmoidal Function
7. Interpretation of Sigmoid Function Results
8. Ignoring the time factor

Fuzzy Cognitive Maps (5/6)

To enhance the knowledge of the system we divide the Concepts of a FCM into the following 3 categories:

1. State Concepts: Concepts that describe the operation of the System, x.
2. Input Concepts: System Inputs, u.
3. Output Concepts: Concepts that describe the outputs of the System, y.

​

This way we have a better knowledge of the system. The proposed separation of Concepts facilitates not only an understanding of the operation of the System but also the calculation of Concepts prices.

Fuzzy Cognitive Maps (6/6)

1. Adjustment of Input Concepts

Defuzzification with membership function (CoA) �

4) Interpretation of Results

�Fuzzification of Output’s value through the trapezoid membership function

Case Study (1/17)

• The new product under consideration in this thesis is the pedalec bike Ideal Orama by Nikos Maniatopoulos SA.
• This company is based in Patras and specializes in designing, producing and selling bicycles. Since 2015 it has assembled electric bicycles, in 2019 for the first time it will export more electric bikes than conventional bicycles, while its 5-year plan is dominated by electric assistance on bicycles of daily use, racing, cargo, rental and sharing. .
• The aim is to use a FCM in the design of the above bike, to make a first assessment of whether the company is worth investing in.
• Our system will consist of 3 Input Concepts, 4 State Concepts and 1 Output Concept.

Case Study (2/17)

• Input Concepts
• Research and Development Cost (R&D cost): C1
• Design: C2
• Quality and Reliability: C3
• State Concepts
• Retail Price: C4
• Cost of Use: C5
• User Friendliness: C6
• Connectivity: C7

Case Study (3/17)

Output Concept

P.P.D (Product Planning Decision): C8

The "decision" of the Product Design System. With the help of the Trapezoidal Participation Function we define the possible outputs (where they range in [0,1]) as follows:

  p.p.d ∈ [0, 0.25]: << Kill the Project >>

  p.p.d ∈ (0.25, 0.5]: << Reconsider Specs >>

  p.p.d ∈ (0.5, 0.75]: << Proceed with the project cautiously >>

  p.p.d ∈ (0.75, 1]: << Go for it >>

Case Study (4/17)

Output’s Membership Function

Case Study (5/17)

Identifying Interconnections between the Concepts

In order to implement our system using the new mathematical Model that calculates the new value of each Concept for each iteration. It is necessary to first form the Weight Matrix and initialize the parameters. To this end we have given the system to 2 Experts working on IDEAL BIKES to form the Linguistic Matrix of interconnections between Concepts.

The Linguistic values ​​used by the Experts will have the following meaning:

• Z (zero) means that the nodes Ci and Cj are not related to each other (denoted by Z in the tables).
• W (weak) which means that the relationship between node Ci and Cj is weak.
• M (medium) which means that the relationship between node Ci and Cj is moderate.
• S (strong) which means that the relationship between node Ci and Cj is strong.
• VS (very strong) which means that the relationship between node Ci and Cj is very strong.
• However, in addition to these Linguistic values, experts also determine the sign of the relationship. (positive (P) or negative (N))

Case Study (6/17)

Linguistic Matrix for the 1^st expert

1^st Expert: Sales & Marketing Director

Case Study (7/17)

2^nd Expert: Product Manager Director

Linguistic Matrix for the 2^nd expert

Case Study (8/17)

Defuzzification of Linguistic Values

• In order to extract the weight matrix, all linguistic values must be converted to numerical. This is achieved through defuzzification.
• The defuzzification technique to be used is the COA-Center of Area method.

Case Study (9/17)

Weight Matrix

Case Study (10/17)

FCM Model

Case Study (11/17)

A.

B.

C.

D.

And result in individual weight matrices

Case Study (12/17)

• For each product separately the Experts should provide word values for all input and status concepts.
• With the help of the Trapezoidal Participation Function we will export the starting values we will enter into our code for each Concept.

Case Study (13/17)

• Applying the new approach of the Mathematical Model we analyzed earlier we will first check whether the MATLAB model we created with the help of MATLAB leads to the expected output (p.p.d) for 2 Ideal bikes already on the market.
• The first bike is the CITYRUN which sold out and is therefore considered successful as a product and the second is the TRAXER-E9 whose sales were not as expected.
•  Therefore in the first bike the expected output (p.p.d) should be in the interval [0.5-1] and in the second in the interval [0-0.5].
• Then the new FCM model will be implemented in the version of the ORAMA model designed for the future.

Case Study (14/17)

For CITYRUN, Expert-based Concepts initial values are:

• C1 = 0.4
• C2 = 0.4
• C3 = 0.9
• C4 = 0.5
• C5 = 0.3
• C6 = 0.4
• C7 = 0.2

And after running the code, the Output C8 = 0.9008 where it is in the interval (0.75-1] => << Go for it! >>

Case Study (15/17)

For the TRAXER-E9, the Expert-Based Concepts initial values are:

• C1 = 0.9
• C2 = 0.8
• C3 = 0.8
• C4 = 0.9
• C5 = 0.3
• C6 = 0.2
• C7 = 0.5

And after running the code, the Output C8 = 0.4941 where it is in the interval (0.25-0.5] => << Reconsider Specs >>

Case Study (16/17)

For the ORAMA the Expert-Based Concepts initial values are:

• C1 = 0.9
• C2 = 0.7
• C3 = 0.9
• C4 = 0.9
• C5 = 0.5
• C6 = 0.3
• C7 = 0.6

And after running the code, the output C8 = 0.7283 where it is in the interval (0.5-0.75] => << Proceed with the Project Cautiously >>

Case Study (17/17)

• We therefore conclude that while this product is worth promoting in the market, changing some of its technical features (cost reduction or increased connectivity) may have further enhanced its sales success.

​

• Experts can, through the system we have created, 'play' with different scenarios to make sure that their choices maximize System Output (interval (0.75-1)) for ORAMA to become a low-risk, successful investment for the company .

Comments - Conclusions

• Our aim was to investigate the extent to which Fuzzy Cognitive Maps could contribute to the study of an existing Business process such as that of Product Planning. The case studies that we ran gave satisfactory conclusions and we can say that the model gives a quick and dirty first look at the company for the prospect of new bike models.
• In this work, the new FCM model is used which gives more flexibility and more accurate output values, with less iterations. The expert can easily modify the system parameters and study their effect on the output of the system.

Future Research

• Using the new FCM model, several of its advantages were verified. However, it would be interesting to study this system, taking into account even more and less obvious possible parameters that influence the Output’s result. In other words, increase the number of Concepts to make the Output of the system even more representative.
• In addition, we could test this model in another company, with a larger number of experts to see how the system would respond in this case.

QUESTIONS??

THANK YOU FOR

YOUR ATTENTION

​

Nikolaos Zervos and Peter P. Groumpos

University of Patras, Greece.

For more information groumpos@ece.upatras.gr

​