RULE-BASED EXPERT SYSTEMS
- Expert systems represent knowledge about a particular field like medicine, business, engineering etc.
- An expert system is any program that functions as an expert.
- The programs which can solve problems by bringing body knowledge are called knowledge-based system or expert systems.
- The knowledge that is used refers to knowledge of experts instead of referring to knowledge acquired from books or non-experts.
- The expert systems basic structure is shown in below figure.
Fig: Basic Structure if an Expert System
- Knowledge base and inference engine are main parts of the system. These parts of the system are explained as follows,
- Inference Engine: The requested information by the user is deduced by it using some processes like forward chaining or backward chaining to perform manipulation on the knowledge base.
- User Interface: It may allow the user to interact with the system by using some kind of natural language processing system it might contain. Graphical user interfaces containing menus may also be used.
- Explanation Subsystem: This component analyzes the structure of reasoning performed and helps in explaining it to the user.
- Knowledge Base: The predicate calculus facts and rules about the subject which is used are present in it.
- Knowledge Acquisition Subsystem: This helps the knowledge engineer while constructing system who works to represent the knowledge in a form, so that it can be entered into knowledge base by discovering any inconsistencies in the knowledge base which is growing and also checks if any incomplete information is present. If such consistencies occur they are reported to the expert for resolution.
- The process of constructing the knowledge base system go through many cycles.
- At every cycle a model is built to test whether it reaches the expected results of experts, if it does not reach the expected results the explanation subsystem is used to know the discrepancies and they are corrected and the process is continue till the satisfied results are achieved.
- The expert systems that are based on reasoning with propositional logic horn clauses are called rule-based expert systems.
- In rule-based-expert systems the rules which are collected from experts are present in knowledge-bases.
- The example of where rule based expert systems are used is explained here,
- Example: The example is to help in making decisions to sanction a loan or not in a bank by using a system.
- Loan approval system has many factors to be considered. Let us take some of the factors to make the example simple.
- The atoms to denote the propositions used here are as
- BALS (The person who applies for the loan has an excellent balance sheet).
- INCE (The income should exceed his/her expenses).
- COLLATE, (The collateral for the loan should be satisfactorily).
- OK (The loan can be sanctioned).
- REP (The person who applies for a loan should have good financial reputation).
- PYMT (The applicant is able to repay loan).
- VAL (The value of collateral or field must be greater than, the amount of loan).
- RATING (A good credit rating must be there for the applicant)
- FARMER (The applicant is a farmer)
- FIE (The applicant has a field)
The rules that are used to take a decision are,
- FARMER ∧ COLLATE ⊃ OK
- COLLATE ∧ PYMT ∧ REP ∧ OK
- RATING ⊃ REP
- INC ⊃ PYMT
- BALS ∧ REP OK
- VAL ⊃ COLLATE
- FIE → FARMER
- If for an applicant the loan officer can check to sanction the loan by checking whether OK is true or not.
- AND/OR proof tree is used to denote OK as root node and the rules which are true as leaf nodes.
- Backward-chaining or forward chaining can be used to know whether OK is true or not.
- Here we use backward-chaining
- The loan can be sanctioned if the applicant is a farmer and has a field that has a satisfactory value i.e., COLLAT has to be proved to be true.
- It can also be sanctioned if both BALS and REP are proved to be true or COLLATE, PYMT and REP are proved to be true.
- These three ways can be represented by the three OK nodes which can be represented under the node OK.
- These three become the OR nodes in AND-OR tree and the result can be proved by proving any one of the nodes.
- The search tree is shown in the below figure.
Figure: Search Tree for a Proof
- It is assumed that the trueness of falsity of RATING, VAL) FIE, INCS and BALS can be available by querying the databases.
- But the information about OK, COLLATE, PYMT, FARMER, REP information is not known to officer nor present in database.
- So these nodes can be linked by the rules.
- In depth-first process so that if any of the leaf nodes like BAL is not true then the search may continue using Rule 1 or Rule 2.
- If the values of INCS, RATING, VAL, FIE are known then we can have a proof for OK.
- The search tree will be broader in realistic applications.
- To get to askable questions backward chaining is used.
- From the proof tree which is built forward chaining is used to query. the consulting systems to get the truth values of the propositions to draw a conclusion.
The system responses for the queries are as follows:
User: Why do you think the loan should be sanctioned?
System: Because the person is a farmer and the collateral for the loan is satisfactory.
User: Why do you think that the Collateral of loan is satisfactory?
System: Because the value of the collateral is greater than the loan amount.
User: Why did you ask me about whether the applicant has a field.
System: Because I am trying to establish whether or not he is .a farmer.
User: Why are you concerned about whether he is a farmer or not?
System: Because it is one of the criteria in establishing whether the loan should be sanctioned or not.
User: Why are you trying to establish whether or not the loan should be sanctioned?
System: Because you asked me to do so.
The users may not answers to questions correctly because of the uncertain rules to which the system have access in many applications.
