Decision Support System and Expert System

DSS framework. A Decision Support System (DSS) is an interactive computer-based information system that helps managers make decisions on semi-structured and unstructured problems, using data, models and a friendly interface. The classic framework (Gorry and Scott Morton, extended by Sprague and Carlson) has two aspects: the problem classification and the components of the DSS.

1. Problem framework (Gorry and Scott Morton). Decisions are classified on two axes.

• Type of problem: structured, semi-structured and unstructured.
• Level of management activity (Anthony): operational control, management control and strategic planning.

This gives a matrix in which structured/operational decisions (e.g. inventory reordering) suit transaction systems and MIS, while unstructured/strategic decisions (e.g. entering a new market) need DSS and managerial judgement.

2. Components of a DSS (Sprague framework).

• Data management subsystem: a database (internal data from transactions, external data from outside sources, and data warehouse) with a database management system and a query facility for retrieving and updating data.
• Model management subsystem: a model base holding financial, statistical, forecasting, optimization and simulation models, with a model base management system (MBMS) for building, storing, running, integrating and updating models.
• Knowledge-based management subsystem: provides expert knowledge and intelligence to support the other subsystems (e.g. rules and an inference engine).
• User interface (dialog) subsystem: lets the decision-maker communicate with the system, using menus, graphics, natural language and reports. It is the part users see and is key to acceptance.
• The user (decision maker): manager or analyst, sometimes assisted by staff.

3. Technology levels. Specific DSS (the application that solves a problem), DSS generator (software that builds specific DSS quickly) and DSS tools (hardware and software used to build generators).

4. Development framework. Includes the people involved: the user, the builder, the intermediary and the toolsmith, with the iterative development process.

Characteristics. Supports semi-structured problems, supports all levels of management, is flexible and adaptive, interactive, uses models, focuses on effectiveness more than efficiency, and gives "what-if" and sensitivity analysis.

Types. Data-driven, model-driven, knowledge-driven, document-driven, communication-driven (group) DSS.

Decision-oriented diagnosis approach. This is an approach to DSS development which starts from the decisions being made and studies how they are made, before any design of the system. Instead of starting from available data or technology, the developer first diagnoses the decision process and the needs of the decision-maker, and then designs a DSS to support them. It is associated with Stabell and with the Gorry and Scott Morton line of analysis, and with the DSS research on how managers decide.

Steps of the diagnosis.

1. Identify the decision and the decision-maker. Who decides, what is the decision, how often, how important it is, and what the context is.
2. Describe the existing decision process. Study how decisions are actually made: the stages (Simon's intelligence, design, choice, and review), the information used, the models or rules of thumb employed, and the role of judgement.
3. Classify the decision. Structured, semi-structured or unstructured; operational, tactical or strategic; individual or group. This determines how much support is possible.
4. Diagnose problems in the present process. For example missing or late information, poor analysis, inability to examine alternatives, conflicts among decision-makers, or cognitive biases.
5. Identify needed support. Determine what information, models, analytical tools, data access and interface are required for each stage of the decision.
6. Specify the DSS design. Choose the type of DSS and its components (data, models, knowledge, dialog) and the form of support (data retrieval, what-if analysis, simulation, optimization or recommendation).
7. Implement, evaluate and refine. Build prototypes iteratively with user feedback, since needs change as users learn.

Features of the approach.

• It focuses on the decision and the users and their cognitive style, instead of technology.
• It treats development as an adaptive, evolutionary and iterative process.
• It involves users heavily, which increases acceptance.
• It supports and improves human judgement, instead of replacing it.

Advantages. The DSS fits real needs, avoids building unused features, and improves decision effectiveness. Limitations. It needs much analyst time and good cooperation of decision-makers, and decision processes may be hard to describe or may change.

Importance of GDSS

A Group Decision Support System (GDSS) is an interactive, computer-based system that helps a group of decision makers work together on semi-structured and unstructured problems. Important decisions in organisations are rarely taken by one person, so a GDSS is important because:

• Better quality decisions: the pooled knowledge, experience and viewpoints of many members reduce the chance of overlooking alternatives.
• Parallel participation: all members can enter ideas at the same time, so no one is blocked by a dominant speaker (reduced production blocking).
• Anonymity: members can contribute without fear of criticism or of the status of the speaker influencing the idea.
• Reduced meeting time and cost: structured agendas, voting tools and automatic recording of minutes shorten meetings.
• Support for distributed groups: members in different places and at different times can still participate.
• Reduced group problems: it lessens groupthink, domination, free-riding and information overload.
• Permanent record: all ideas, votes and results are stored for later reference and consistency.

Types of group decision support systems

GDSS are classified by the place and time of the group's work.

1. Decision room (face-to-face, same time, same place): members meet in one room equipped with a networked workstation for each participant, a public display screen and GDSS software. Used for brainstorming, idea ranking and voting. Anonymous input is possible.
2. Legislative session: members work in the same place but over an extended period. A facilitator and a large screen support a debate in which members present arguments in turn; the system records and organises the contributions.
3. Local area decision network: members are in the same place (building) but work at different times, each on his or her own workstation connected over a LAN. Messages and decisions are stored on a shared server. Suitable when members are busy at different times.
4. Teleconferencing / virtual meeting (same time, different place): members at distant locations meet through audio/video links and shared screens, supported by decision tools such as voting and shared whiteboards.
5. Computer-mediated conferencing (different time, different place): also called a wide-area decision network. Members contribute through e-mail, discussion boards and shared documents at their own convenience. Useful for geographically dispersed teams and long-running decisions.

Typical GDSS software tools: electronic brainstorming, idea organiser, issue analyser, stakeholder identification, questionnaire and voting tools, and a group dictionary.

A Transaction Processing System (TPS) records and processes the routine, day-to-day transactions of an organisation (sales, payroll, orders, bank deposits). A Decision Support System (DSS) is an interactive system that helps managers analyse data and models to make semi-structured or unstructured decisions.

The TPS is usually the main source of data for the DSS: the TPS captures data, and the DSS extracts and analyses it.

Building a DSS is a project involving several groups of people, each with a distinct role.

1. Manager / end user: the person who will use the DSS to make decisions. He or she defines the decision problem and the requirements, tests the prototype and gives feedback. User involvement is the key factor in the success of a DSS.
2. Champion (sponsor): a senior person who supports the project, secures funds and management commitment, and removes organisational obstacles.
3. DSS builder (developer / designer): the technical person who designs, programs and integrates the DSS components (database, model base, user interface). He or she may be a systems analyst or programmer.
4. Intermediary (user support person): works between the user and the builder, helps the user to operate the system, interprets requests and explains outputs. This is also called a staff assistant or facilitator.
5. Technical supporter: develops additional or specialised capabilities such as new models, databases, tools or interfaces which the builder uses.
6. Systems / tool builder: develops the DSS generators and tools (software packages, languages) used to build specific DSS.
7. Toolsmith / database administrator: maintains the data resources, provides access to data, and keeps data secure and consistent.
8. Expert / domain specialist: supplies knowledge of the problem area, models and business rules.

In small projects one person may play several of these roles.

Operating data is the detailed data produced and used by the day-to-day transaction systems (TPS). DSS data is the data that has been selected, extracted and organised from operating and external sources to support analysis and decisions.

DSS data is usually stored in a data warehouse or data mart, populated from the operating data by extraction, transformation and loading (ETL).

An expert system (ES) is a computer program that captures the knowledge of human experts in a narrow domain and uses it to solve problems and give advice as an expert would.

Components of the architecture

1. Knowledge base: contains the domain knowledge as facts and heuristic rules (usually IF-THEN production rules, also frames or semantic networks). It is the heart of the system and is separate from the inference procedure.
2. Inference engine: the reasoning mechanism. It matches facts in working memory with rules in the knowledge base and derives new facts or conclusions. Its strategies are:
   • Forward chaining (data driven): starts from known facts and fires rules until a goal is reached.
   • Backward chaining (goal driven): starts from a hypothesis and searches for facts supporting it. It also has a conflict-resolution strategy when several rules can fire.
3. Working memory (blackboard): a temporary store of the facts of the current problem, user inputs and intermediate conclusions.
4. User interface: lets the user enter the problem data and view results, usually in a question-and-answer or menu dialogue in natural language or graphics.
5. Explanation subsystem: explains why a question is asked and how a conclusion was reached, which builds the user's trust and helps debugging.
6. Knowledge acquisition subsystem: helps the knowledge engineer to collect, encode and update knowledge from experts, books and databases.

Flow: the user states the problem through the interface, the inference engine uses the knowledge base and working memory to reason, and the answer with an explanation is returned to the user. The domain expert and knowledge engineer build the knowledge base, which makes the system easy to modify because knowledge is separated from control.

Fuzzy reasoning (approximate reasoning) draws conclusions from fuzzy rules and imprecise inputs using fuzzy set theory. It is performed in four main steps, as in a Mamdani fuzzy inference system.

1. Fuzzification: each crisp input is converted into degrees of membership in linguistic sets using membership functions. For example, a temperature of 28 degrees may be "warm" with membership 0.7 and "hot" with membership 0.2.
2. Rule evaluation (inference): the fuzzy rule base of IF-THEN rules is applied, e.g. IF temperature is hot AND humidity is high THEN fan speed is fast.
   • The antecedent strength is found with fuzzy operators: AND as the minimum, $\mu_{A\wedge B}=\min(\mu_A,\mu_B)$, and OR as the maximum, $\mu_{A\vee B}=\max(\mu_A,\mu_B)$; NOT is $1-\mu_A$.
   • The strength is applied to the consequent (clipping by min or scaling by product).
3. Aggregation: the output fuzzy sets of all fired rules are combined (usually by union/maximum) into one fuzzy set for each output variable.
4. Defuzzification: the aggregated fuzzy set is converted into a single crisp value. The common centroid method is $z^{*}=\frac{\int z\,\mu(z)\,dz}{\int \mu(z)\,dz}$ Other methods are mean of maximum and weighted average.

The result is a crisp, actionable output, such as a fan speed or a loan risk score, derived from vague human-like rules.

The strategic impact grid (McFarlan's strategic grid) is a framework that classifies an organisation's information systems according to their impact on the business, to decide how much attention and investment IT deserves. It has two dimensions:

• Vertical axis: strategic impact of the existing (current) operating information systems, low to high.
• Horizontal axis: strategic impact of the planned (future) application development portfolio, low to high.

This gives four quadrants:

1. Support: low impact now and in future. IT only supports routine operations (e.g. payroll, word processing). Emphasis is on cost control; IT is not critical.
2. Factory: high impact of current systems but low impact of future applications. The business depends on reliable operation of existing systems (e.g. airline reservation, manufacturing control), so stress is on reliability, efficiency and uptime.
3. Turnaround: low impact of current systems but high expected impact of new applications. IT is about to become critical; the firm needs to evaluate and select promising new technologies and manage risk.
4. Strategic: high impact now and in the future. IT is central to competitive advantage and survival (e.g. banking, e-commerce); top management must be closely involved in IT planning.

The grid helps management place IT, set priorities, choose the level of executive involvement and allocate resources. A firm may move between quadrants over time, and DSS or expert systems may be strategic in one quadrant and merely support in another.

The expert system development life cycle (ESDLC) is the sequence of phases for building an expert system. It differs from the conventional SDLC because knowledge must be acquired from experts and development is iterative with prototyping.

1. Problem identification and feasibility study: decide whether the problem suits an expert system (a narrow domain, knowledge available, experts cooperative, payoff justifies the cost). Study technical, economic and operational feasibility.
2. Project planning and requirements: define scope, objectives, users, resources, team (domain expert, knowledge engineer, programmers) and schedule.
3. Knowledge acquisition: the knowledge engineer collects knowledge from the domain expert (interviews, observation, case analysis) and from documents. This is usually the bottleneck.
4. Knowledge representation and design: knowledge is organised and encoded as rules, frames or semantic nets. The architecture, inference strategy (forward or backward chaining) and interface are designed, and the development tool (shell or language such as Prolog, CLIPS) is chosen.
5. Prototype development and implementation: a small working prototype is built, the knowledge base is expanded step by step and the system is refined.
6. Testing, verification and validation: verification checks that the system is built correctly (no conflicting, redundant or circular rules); validation checks that its answers match those of experts on test cases.
7. Deployment and training: the system is installed, integrated into the working environment and users are trained.
8. Maintenance and evolution: knowledge is updated as the domain changes, errors corrected and new features added.

The phases are cyclic: feedback from testing returns to knowledge acquisition and design.

A fuzzy expert system combines the rule-based reasoning of an expert system with fuzzy logic so that it can handle imprecise, vague knowledge expressed in linguistic terms. The main types are:

1. Mamdani-type fuzzy expert system: both antecedents and consequents of the rules are fuzzy sets, e.g. IF speed is high THEN braking is strong. The aggregated fuzzy output is defuzzified (e.g. by centroid). It is intuitive and widely used in control, but computationally heavier.
2. Sugeno (Takagi-Sugeno-Kang) type: the consequent is a crisp function of the inputs, typically a constant or linear function: IF $x$ is $A$ AND $y$ is $B$ THEN $z=px+qy+r$. The output is a weighted average of the rule outputs, $z=\dfrac{\sum w_i z_i}{\sum w_i}$, so no defuzzification is needed. It is efficient and suits adaptive and optimisation methods.
3. Tsukamoto type: the consequent is a fuzzy set with a monotonic membership function; each rule yields a crisp output from its firing strength, and the final output is the weighted average.
4. Neuro-fuzzy expert system: a neural network learns and tunes the membership functions and rules of the fuzzy system from data (e.g. ANFIS), combining learning with human-readable rules.
5. Genetic-fuzzy / evolutionary fuzzy system: a genetic algorithm optimises the rule base and membership functions.
6. Fuzzy expert systems with certainty factors / hybrid systems: rule uncertainty is expressed together with fuzziness for diagnosis and decision support.

Uses include medical diagnosis, industrial control, credit scoring and weather prediction.

(a) Cost-Effectiveness Analysis

Cost-effectiveness analysis (CEA) is a technique used to evaluate a DSS (or any project) by comparing its costs with its effectiveness, when the benefits cannot easily be expressed in money. Instead of converting benefits into Rs, effectiveness is measured in natural units such as decision time saved, number of correct decisions, accuracy or customer satisfaction.

• Costs include development, hardware, software, training, operation and maintenance.
• Alternatives are compared by the cost-effectiveness ratio $\text{CER}=\frac{\text{Cost}}{\text{Units of effectiveness}}$ and the option with the lowest cost per unit of effect (or the highest effect for a fixed budget) is chosen.
• It is useful for DSS because many benefits are intangible, such as better quality decisions and faster response.

Its limitation is that it cannot show whether a project is worth doing at all, only which alternative is better.

(b) Executive DSS (EDSS / Executive Information System)

An Executive DSS is a system designed for senior executives to monitor the overall performance of the organisation and support strategic decisions.

• Features: highly user-friendly graphical interface, drill-down from summary to detail, exception reporting, key performance indicators, dashboards, trend analysis, access to internal and external data (competitors, markets, news), and easy use without training.
• Data: summarised, from corporate databases, data warehouse and external sources.
• Uses: strategic planning, performance monitoring, identification of problems and opportunities.
• Benefits: rapid access to key information, better strategic control and focus on critical success factors.

DSS can be classified by the dominant component that drives the support it gives (Power's taxonomy).

(a) Data-driven DSS

• Emphasises access to and manipulation of large volumes of internal and external time-series data.
• Tools: file systems, data warehouses, OLAP, data mining, executive information systems.
• Example: a sales-analysis system that drills down from yearly to monthly sales by region and product.

(b) Model-driven DSS

• Emphasises access to and manipulation of financial, optimisation or simulation models.
• Uses limited data and parameters supplied by the decision maker.
• Tools: what-if analysis, linear programming, decision trees, forecasting.
• Example: a production-planning or loan-evaluation model.

(c) Knowledge-driven DSS

• Provides specialised problem-solving expertise stored as facts, rules and procedures.
• Suggests or recommends actions; related to expert systems and data mining.
• Example: a credit-approval advisor that applies lending rules.

(d) Document-driven DSS

• Retrieves, searches and analyses unstructured documents such as reports, policies, e-mails, web pages and multimedia.
• Tools: search engines, text-mining and document-management systems.
• Example: a system that searches past tender documents and contracts.

Many real systems are hybrids combining two or more of these types.

An enterprise-wide DSS serves many users across departments and locations, so it depends on networks. The main networking issues are:

1. Connectivity and architecture: Linking LANs, WANs and the Internet/intranet across branches, and choosing between client-server, web-based or distributed layouts.
2. Bandwidth and performance: Large data transfers from the data warehouse and OLAP queries can cause delays and response-time problems, especially over slow links.
3. Security: Sensitive strategic data must be protected from unauthorised access, interception and attacks, using authentication, encryption, firewalls and VPNs.
4. Reliability and availability: Network failure stops decision making, so redundancy, backups and fault tolerance are needed.
5. Data integration and consistency: Data stored on heterogeneous platforms must be synchronised and kept consistent across sites.
6. Standards and interoperability: Different hardware, operating systems and protocols require common standards (TCP/IP, HTTP, XML) so that components work together.
7. Scalability: The network must handle growth in users, data volume and new applications.
8. Access and remote use: Mobile and remote users need secure access from anywhere, which raises device and bandwidth issues.
9. Cost and management: Installing, monitoring and maintaining the network, and training staff, are costly.
10. Distributed processing and control: Deciding where data and models reside and who controls them leads to coordination and ownership problems.

A web-based DSS delivers decision support to users through a web browser over the Internet or an intranet.

Three-tier architecture

1. Client tier (presentation): A web browser (PC, tablet or phone) displays HTML/JavaScript pages. The user enters queries and parameters and views reports and charts. No special software is needed on the client.
2. Application/web server tier (logic): The web server receives HTTP requests and passes them to the application server. This tier runs DSS software: the model management system, the dialog (user interface) management, security and session control. It uses scripts or servlets (PHP, JSP, ASP) to build dynamic pages.
3. Data tier: A database or data warehouse, with the knowledge base and model base, that supplies data to the application server.

Request flow $\text{Browser} \rightarrow \text{Web server} \rightarrow \text{Application/DSS server} \rightarrow \text{Database/Model base} \rightarrow \text{back to browser}$

Features and advantages

• Platform independence and easy deployment, with updates only on the server.
• Wide, remote and low-cost access; supports collaboration.
• Easy to integrate with the enterprise intranet, extranet and external data.

Limitations: Dependence on network speed, security risks, and limited interactivity compared with desktop systems.

Importance of DSS

• Supports semi-structured and unstructured decisions that routine systems cannot handle.
• Improves the speed and quality of decisions by quickly analysing large amounts of data.
• Allows what-if and sensitivity analysis to compare alternatives and reduce risk.
• Helps managers handle complexity, changing business conditions and competition.
• Improves communication, collaboration and cost control, and gives a competitive advantage.

Characteristics of DSS

• Supports semi-structured and unstructured problems.
• Supports all managerial levels and groups, and individual or group decisions.
• Interactive, with a friendly user interface and quick response.
• Adaptable and flexible, and can be changed as needs change.
• Uses data and models together (database, model base, user interface).
• Supports the whole decision process: intelligence, design and choice.
• Controlled by the user, who can run what-if analysis and sensitivity analysis.
• Improves effectiveness of decisions rather than only efficiency, and supports rather than replaces the decision maker.

Structured vs unstructured decisions

(Semi-structured decisions lie between the two, with part of the problem structured.)

How DSS helps in decision making

• Gives quick access to relevant data and information.
• Applies models for forecasting, simulation and optimisation.
• Allows what-if and sensitivity analysis to compare alternatives.
• Supports the intelligence, design and choice phases, mainly for semi-structured and unstructured decisions, by combining the judgment of the manager with computer analysis.
• Improves the speed, accuracy and consistency of decisions and supports group work.

The user interface (dialog component) decides how easily the user can use the DSS. The main factors in its design are:

1. User characteristics: Knowledge, skills, experience and cognitive style of the users (managers, analysts, novices).
2. Task requirements: The kind of decision, information needed and output format (tables, graphs, reports).
3. Interaction style: Menus, command language, forms, natural language, graphical (GUI) and direct manipulation, chosen to fit the user.
4. Ease of use and consistency: Simple, clear, consistent screens, with help and error messages.
5. Presentation of information: Proper use of graphics, colour, layout and visualisation (charts, dashboards).
6. Flexibility: Different views and options for different users, and support for what-if analysis.
7. Response time and performance: Fast feedback to keep the user engaged.
8. Hardware and software environment: Available devices, display, network and web/mobile access.
9. Feedback and user involvement: Prototyping, testing with users and iterative improvement.
10. Security and access control: Role-based access.

Data mining tools and techniques

• Tools: WEKA, RapidMiner, KNIME, Orange, SAS Enterprise Miner, IBM SPSS Modeler, R and Python libraries.
• Techniques: classification (decision trees, naive Bayes, SVM), clustering (k-means, hierarchical), association rule mining (Apriori), regression, neural networks, genetic algorithms, nearest-neighbour, anomaly detection and case-based reasoning.

Case-based reasoning (CBR)

CBR solves a new problem by reusing the solutions of similar past cases stored in a case base.

The CBR cycle (4 Rs):

1. Retrieve: Find the most similar cases using a similarity measure such as nearest-neighbour matching.
2. Reuse: Apply the solution of the retrieved case to the new problem.
3. Revise: Adapt and test the proposed solution if it does not fit exactly.
4. Retain: Store the new problem and its verified solution as a new case.

Example: a help desk finds an earlier, similar fault ticket and reuses its fix.

Advantages: no explicit rules needed, learns over time, easy to explain. Limitation: depends on a good case base and the similarity measure.

Errors can enter an expert system at any stage of its development. Main sources:

1. Knowledge acquisition errors: The expert may be unable or unwilling to express knowledge, give incomplete, inconsistent or wrong knowledge, or the knowledge engineer may misunderstand it.
2. Knowledge representation errors: Incorrect or ambiguous rules, wrong choice of representation (rules, frames), missing rules, redundant or conflicting rules.
3. Inference engine errors: Faulty reasoning strategy, wrong conflict resolution and incorrect handling of uncertainty.
4. Domain/expert errors: Experts themselves are biased or wrong, and domain knowledge may change or become outdated.
5. Implementation errors: Programming bugs, coding mistakes and wrong tools.
6. Testing and validation errors: Inadequate test cases and poor verification.
7. User interface and data errors: Wrong input data from users, poor questions and misinterpreted outputs.
8. Communication errors: Poor communication between the expert, knowledge engineer and developers.

Fuzzy rule

A fuzzy rule is an IF-THEN rule whose conditions and conclusions use linguistic variables with fuzzy sets rather than crisp values.

Example: IF temperature is high AND humidity is low THEN fan speed is fast.

Here "high", "low" and "fast" are fuzzy sets with membership functions $\mu(x)\in[0,1]$.

Fuzzy reasoning in an expert system

1. Fuzzification: Convert crisp inputs into degrees of membership in fuzzy sets (for example, 32 °C is high to degree 0.7).
2. Rule evaluation (inference): Apply the rules. AND is usually taken as the minimum, OR as the maximum: $\mu_{A\land B}=\min(\mu_A,\mu_B),\qquad \mu_{A\lor B}=\max(\mu_A,\mu_B)$ The result clips (Mamdani) or scales the output fuzzy set of each rule.
3. Aggregation: Combine the outputs of all rules into a single fuzzy set.
4. Defuzzification: Convert the aggregated fuzzy set into a crisp output, commonly by the centroid method: $x^*=\frac{\int x\,\mu(x)\,dx}{\int \mu(x)\,dx}$

Fuzzy expert systems are mainly of the following types, according to how rules and outputs are formed.

1. Mamdani-type fuzzy expert system

• Both antecedent and consequent are fuzzy sets.
• Output is a fuzzy set, which must be defuzzified (centroid etc.) to get a crisp value.
• Intuitive and widely used, but computationally heavier.
• Example rule: IF error is large THEN valve opening is wide.

2. Sugeno (Takagi-Sugeno-Kang) type

• Antecedent is fuzzy, but the consequent is a crisp function, a constant or linear equation, e.g. $z = ax + by + c$.
• Output is a weighted average of rule outputs, so no separate defuzzification is needed.
• Computationally efficient; suitable for control and optimisation.

3. Tsukamoto type

• Consequent is a fuzzy set with a monotonic membership function; each rule gives a crisp output and the final output is a weighted average.

4. Hybrid systems (neuro-fuzzy and genetic-fuzzy)

• Combine fuzzy logic with neural networks (e.g. ANFIS) or genetic algorithms so that the membership functions and rules can be learned or tuned automatically.

Fuzzy systems may also be classified as pure fuzzy systems (fuzzifier, rule base, inference engine, defuzzifier) and systems with fuzzifier/defuzzifier connecting crisp inputs and outputs to the fuzzy core.

Several kinds of persons interact with an expert system:

1. Domain expert: The human specialist whose knowledge and problem-solving skill are captured in the system.
2. Knowledge engineer: Elicits knowledge from the expert, represents it in a suitable form (rules, frames) and builds the knowledge base.
3. System builder/programmer: Designs and implements the software, the inference engine, the user interface and the tools.
4. User (end user): The person who consults the system to solve a problem or get advice, such as a non-expert, a client, a manager or a student.
5. Project manager: Plans, coordinates and controls development.
6. Maintenance personnel/administrator: Updates the knowledge base, corrects errors and maintains the system after delivery.
7. Tester/validator and clerical staff: Test the accuracy of the system and enter data.

(a) ROMC Design Approach

ROMC (Representations, Operations, Memory aids, Control mechanisms), proposed by Sprague and Carlson, is a design approach based on the user's view of the DSS rather than on the technical components.

• Representations: The ways information is presented to the user, such as tables, charts, graphs, reports and maps.
• Operations: The actions the user performs on the representations: analyse, compare, select, simulate, forecast, what-if.
• Memory aids: Tools that help the user remember and organise, such as databases, views, workspaces, links and saved results.
• Control mechanisms: Means by which the user controls the system, such as menus, commands, defaults and help.

The approach is used to build a DSS which fits the user's needs through iterative development.

(b) Group Decision Support System (GDSS)

A GDSS is an interactive computer-based system that supports a group of decision makers working together to solve unstructured problems.

• Components: hardware (meeting room, networked PCs, projectors), software (brainstorming, voting, ranking, idea organisation, group database and models), people (group members and a facilitator) and procedures.
• Features: anonymous input, parallel communication, automated record keeping and structured group techniques (Delphi, nominal group).
• Settings: same time/same place (decision room), same time/different place, different time/same place, different time/different place.
• Advantages: more ideas, less dominance by individuals, better participation and faster decisions. Limitations: cost, possible information overload, loss of face-to-face interaction.