applying ai in insurance domain

Upload: kpss007

Post on 09-Apr-2018

217 views

Category:

Documents


0 download

TRANSCRIPT

  • 8/7/2019 Applying AI in Insurance Domain

    1/29

    Applying AI in Insurance Domain

    1 Introduction to Intelligent system:

    The re-engineering of old, existing information systems and their

    transformation in modern, extensible, scalable, viable systems is a complex and tedious

    process involving significant costs and resources.

    1.1Overview of Intelligent Systems :

    Artificial intelligence is a dynamic, varied, growing field. Its applied

    technologies range from expert system to computer vision. In this part we first present an

    overview of expert systems. These systems are constructed knowledge engineering,

    which involves several tasks. First knowledge is collected (from people or from

    documented sources) by a process called knowledge acquisition. Knowledge acqusisition

    can be accomplished manually and with some degree of automation. Then the acquired

    knowledge is organized into a knowledge base. In many systems knowledge

    representation involves IF-THEN rules, but there are other useful representations (such as

    frames).

    Represented knowledge is used through reasoning or inferencing,

    procedures, which can be done through under assumption certainty or uncertainty. Theknowledge engineering development process is described.

    1.2 Expert System:

    1.2.1What is Expert System:

    [1]In this paper use of expert systems speed-up of human

    professional work, human productivity, financial benefits and a better answer to

    users needs. Last decade shows that a growing number of organizations shift their

    informational systems towards a knowledge-based approach. Expert system is

    used in many different fields like stock market advisors, commodity trading,

    financial planning, tax preparation and planning, granting of loans and

    determination of credit limits, diagnosis and treatment of various diseases,

    determination of chemical properties of unknown compounds ,scheduling and

  • 8/7/2019 Applying AI in Insurance Domain

    2/29

    control of automated factory, configuration and design of computers, layout and

    design of printed circuit boards.

    [1] Talked about the following types of benefits from ES:

    (1) Better customer service.

    (2) Reduction in time to complete tasks:

    two weeks to 30 minutes;

    50 man-days to two man-days;

    15 man-days to one man-day.

    (3) Organizational learning increased.

    (4) Increases in production.

    (5) Fewer defects.

    (6) Better quality.

    (7) Shorter time to diagnosis.

    (8) More effective uses of resources.

    (9) More consistent decision making.

    (10) Reduction in emergency calls.

    (11) Reduction in staff.

    (12) Redeployment of personnel.

    Complex decisions involve intricate combination of factual and heuristic

    knowledge. In order for the computer to be able to retrieve and effectively use

    heuristic knowledge, the knowledge must be organized in an easily accessible

    format that distinguishes among data, knowledge, and control structures. Expert

    system is organized in three different levels they are:

    1) Knowledge base consists of problem-solving rules, procedures, and intrinsic

    data relevant to the problem domain.

    2) Working memory refers to task-specific data for the problem under

    consideration.

    3) Inference engine is a generic control mechanism that applies the axiomatic

    knowledge in the knowledge base to the task-specific data to arrive at some

    solution or conclusion.

  • 8/7/2019 Applying AI in Insurance Domain

    3/29

    Expert System Structure: [17]The modularity of an expert system is an

    important distinguishing characteristic compared to a conventional computerprogram. Modularity is affected in an expert system by the use of three distinctcomponents, as shown in Fig 1.

    1.2.2 Development process:

    The development process for ES has changed over the years.

    Originally, all ES were developed in specialized AI languages such as LISP or

    PROLOG. Both of these are languages that require specialized training, even for

    experienced programmers in traditional languages such ignore programming

    concerns and concentrate on structuring the experts knowledge in the system.

    The structure is usually in the form of if-then rules, although other structures

    such as objects are also used. After an initial prototype is developed, an ES

    typically goes through several iterations. These iterations consist of alternately

    testing by an expert or experts and modifications to the system by the developer

    KnowledgeEngineer

    InferenceEngine

    WorkingMemory

    KnowledgeBase

    User Expert

    Database

    Fig 1

  • 8/7/2019 Applying AI in Insurance Domain

    4/29

    using the suggestions of the experts. After the ES reaches an acceptable level of

    performance, the implementation process to the users is started.

    [2] Presents CAPE a powerful tool for building a new generation of

    KBSS. It combines the strengths of two well-established tools with very powerful

    but complementary pattern-matching mechanisms. Perl's ability to search text and

    match powerful regular expressions is unequaled, while CLIPS provides powerful

    mechanisms for finding patterns of combinations of symbolic information. The

    CAPE programmer can exploit the strengths of both, using Perl to analyze

    documents or query results, and CLIPS to recognize and react to the combinations

    of matches found. CAPE provides powerful mechanisms to support a number of

    key activities:

    Symbolic reasoning: CLIPS offers a very efficient forward chaining rule-based

    system with extremely expressive pattern matching, coupled with a highly

    flexible object-oriented system and supported by a truth-maintenance system.

    Data analysis/manipulation: Perl has extremely powerful regular expression

    matching coupled with very concise string handling and easy-to-use hash-based

    index-building and data structuring.

    Service Provision: CAPE'S socket monitoring mechanisms allow a rule-based

    program to remain responsive to external activity even while it is reasoning.

    Standard languages/libraries: CAPE programs can use any of the enormous

    range of software available in CPAN, the Comprehensive Perl Archive Network.

    Interaction: with software packages Perl provides very concise and flexible

    mechanisms for controlling and processing the results obtained from system

    commands and other external programs. CAPE programmers can also exploit the

    tools for generating Perl "wrappers" for software components written in C, and

    make use of Perl's ability to dynamically load compiled code at run-time.

  • 8/7/2019 Applying AI in Insurance Domain

    5/29

  • 8/7/2019 Applying AI in Insurance Domain

    6/29

    CommonKADS, MIKE & PROTIEGE-II

    PSM in KE

    Current developments in KE and their relationships to other disciplines.

    Knowledge Engineering Process:

    [6] talk about knowledge Engineering Process which include five major

    activities:

    Knowledge acquisition: Knowledge acquisition involves the acquisition of

    knowledge from human experts, book, document, sensors or computer files.

    The knowledge may be specific to the problem-solving procedures it may be

    general knowledge or it may be metaknowledge.

    Knowledge validation: The knowledge is validated and verified until its

    quality is acceptable

    Knowledge representation: The acquired knowledge is organized in an

    activity called knowledge representation. This activity involves preparation of

    knowledge map and encoding the knowledge in knowledge base.

    Inferencing: The activities involves the design of software to enable the

    computer to make inference based on the knowledge and specifics the

    problem. Then system can provide advice to a non expert user.

    Explanation and justification: This involves the design and programming of an

    explanation capability, for example programming ability to answer the

    question such as why a specific piece of information is needed by the

    computer orhow a certain conclusion was derived by the computer.

    The process of knowledge engineering and the relationship among these

    activities are shown in fig 2.

  • 8/7/2019 Applying AI in Insurance Domain

    7/29

    KNOWLEDGE ACQUISITION TECHNIQUES:

    Basic techniques used areInterviews:

    The most commonly used form of knowledge acquisition is face-to-face

    interview analysis. It is an explicit and appears in several variation. It involves

    direct dialogs between the expert and the knowledge engineer. Information is

    collected with the aid of conventional instruments and is subsequently

    transcribed, analyzed and coded.

    Unstructured interviews:

    Many knowledge acquisition interviews session are conducted informally,

    usually as a starting point. Starting information save time it helps to move quickly

    to the basic structure of the domain. Usually it is followed by formal technique.

    Source ofknowledge(expert,others)

    KnowledgeValidation (testcases)

    Knowledgerepresentation

    Knowledgebase

    Inferencing

    Explanationjustification

    Fig 2

  • 8/7/2019 Applying AI in Insurance Domain

    8/29

    Unstructured interview provides complete or well-organized description of

    cogenitive process.

    Structured interview:

    Structure interview is a systematic goal oriented process. It forces

    organized communication between expert and knowledge engineer. It reduces

    interpretation problem inherited in unstructured interview, and prevents distortion

    caused by domain expert subjectivity. It is more effective and efficient techniques

    of knowledge acquisition and can be applied to knowledge acquisition from

    multiple experts. While this technique is used experts fill out a set of carefully

    designed questions raised by knowledge engineer making use of established

    domain model of business decision-making activity to capture the subjective and

    qualitative aspects of decision making. Questionnaires can be particularly useful

    in discovering the objects of the domain, in uncovering relationship, and in

    determining uncertainties.

    Observation:

    Sometime it is possible to observe an expert at work. In many ways, this is

    the most obvious and straightforward approach to knowledge acquisition. This

    technique allows an expert to work in accustomed environment without

    interruptions by the knowledge engineer and gives knowledge engineer insights

    into complexities of a problem. Before implementing this method it is necessary

    to decide experts performance recording technique. Recording methods may be

    notes, video etc. The major limitations of

    this technique is that the underlying reasoning in expert s mind is usually not

    reveled in his/her actions.

    Computer-adided approaches:

    The purpose of computerized support for the expert is to reduce or

    eliminate the potential problems. A smart knowledge acquisition tool must be able

    to add knowledge to knowledge base incrementally and refine or even correct

  • 8/7/2019 Applying AI in Insurance Domain

    9/29

    existing knowledge. Benefits derived from using a computer aided environment

    for knowledge acquisition include:

    1) Electronic documentation of knowledge.

    2) Knowledge extraction can be done in parallel from multiple experts.

    3) Conflicts are addressed during knowledge extraction sessions.

    4) Interactions among experts result in an enlarged and enriched domain of

    expertise.

    1.2.2.2 Knowledge representation:

    [7] Presents what is a knowledge representation? We argue

    that the notion can best be understood in terms of five distinct

    roles that it plays, each crucial to the task at hand:

    First, a knowledge representation is most

    Fundamentally a surrogate, a substitute for the thing itself that is

    used to enable an entity to determine consequences by thinking

    rather than acting, that is, by reasoning about the world rather

    than taking action in it.

    Second, it is a set of ontological commitments, that is, an

    answer to the question; in what terms should I think about theworld?

    Third, it is a fragmentary theory of intelligent reasoning

    expressed in terms of three components: (1) the representations

    fundamental conception of intelligent reasoning, (2) the set of

    inferences that the representation sanctions, and (3) the set of

    inferences

    that it recommends.

    Fourth, it is a medium for pragmatically efficient

    computation, that is, the computational environment in which

    thinking is accomplished. One contribution to this pragmatic

    efficiency is supplied by the guidance that a representation

  • 8/7/2019 Applying AI in Insurance Domain

    10/29

    provides for organizing information to facilitate making the

    recommended inferences.

    Fifth, it is a medium of human expression, that is, a

    language in which we say things about the world. Understanding

    the roles and acknowledging.

    [6]book talked about knowledge is represented as IF-THEN or production

    rules. Rule based representation is popular because development shells are

    available easily, shells are less expensive, ease in usability of shell, rules represent

    natural mode of knowledge representation, learning curve for rule-based system is

    much steeper than for any alternative mode of representation, rules are

    transparent, modifications of rules are easier and validation is relatively easy.

    Production rules are normally IF-THEN variety. However, in some instances this

    is extended to include IF-THEN-ELSE rules. The alternate designation for IF-

    THEN rules is that of condition-action; IF this condition (or premise or

    antecedent) occurs, THEN some action (or result or conclusion or consequence)

    will (or should) occurs.

    IF premise, THEN conclusion. If your income is high, THEN your chance of

    being audited by the IRS is high

    Conclusion, IF premise. Your chance of being audited is high, IF your incomeis high

    Inclusion of ELSE. If your income is high THEN your chance of being

    audited is high ELSE your chance of being audited is less.

    More complex rules. If the credit rating is high AND the salary is more

    $30,000 OR assets are more than $75,000 AND pay history is not poor

    THEN approve a loan up to $10,000 and list the loan in category B. The

    action part may include additional information: THEN approve the loan and

    refer the application part may include additional information : THEN approve

    the loan and refer the application to an agent.

    1.2.2.4 Issues in expert systems:

  • 8/7/2019 Applying AI in Insurance Domain

    11/29

    Here in Expert System how different rules get triggered and how they areretrieved are some of the major issues to be studied. Also to it is necessary to see thateach and every rules are covered this is the major part of the Expert system.

    1.3 Case Base Reasoning:

    Expert Advisor works with knowledge acquired from expert and users,

    employing the case base reasoning approach. Once knowledge in the knowledge base is

    completed, or is at least at a sufficiently high of accuracy, it is ready to be used. A

    computer program is needed to access the knowledge for making inferences. This

    program is an algorithm that controls a reasoning process and is usually called the

    inference engine.

    1.3.1 Types of Case Base Reasoning:

    1. Textual CBR: Here in this type of Case Base Reasoning Free text is

    used

    2. Conversational CBR: List of question and answers are used in

    conversational type of Case Base Reasoning. Here no common case

    structure are used

    3. Structural CBR: Database like representation is done in structural Case

    Base Reasoning

    [8] Focuses on case base reasoning system uses the technique to match a

    situation or problem description to a stored database. Here the input is given by the user

    on the current situation and the output is case retrieval to the most similar match to the

    database. The CBR engine first searches for case history that is similar to the given

    description.

    As stated case is a unique knowledge entity describing a problem and

    solution. It can be represented a single database.

    Here representation is:

    A problem point to one or more case.

  • 8/7/2019 Applying AI in Insurance Domain

    12/29

    A case has a single solution.

    A question can influence one or more case.

    CBR systems vary in the way the case database is structured. The

    representation can be flat, where all cases are represented at the same level, or it can be

    hierarchical, expressing relationships between cases and sub-cases. The hierarchical

    organization is useful when the CBR system is used for taxonomic tasks, for example, to

    identify an animal based on its features. A detailed discussion of data structures is beyond

    the scope of this paper, as the structure is not expected to impact the performance of a

    diagnosis system.

    Similarity between Cases:

    It consist of following attribute:

    Reflective: A case is similar to itself.

    Symmetric: If A is similar to B, then B is also similar to A

    Transitive : If A is similar to B and B is similar to C, we cannot say that A is

    similar to C, because the features defining the similarities between A and B and

    between B and C are not necessarily the same.

    CaseProblem

    Case

    Solution

    Solution

    Question

    Fig 3

  • 8/7/2019 Applying AI in Insurance Domain

    13/29

    1.3.2 Knowledge Presentation:

    To calculate similarity we use:

    Numeric: sim(a,b)=|a-b| / Range

    Sumbolic: sim(a,b)= 1 if a=b 0 if a not equal to b

    Multi-valued: sim(a,b)= card(a) card(b)/ card(a b)

    Tazonomy: sim(a,b)= h(common node (a,b))/min(h(a),h(b))

    Where

    Cardis the cardinality (size) of the set

    range is the absolute value of difference between the upper and lower boundary

    of the set .

    h is the height (number of levels) of the taxonomy tree.

    Case Indexing:

    A CBR system its ability to retrieve relevant cases quickly and accurately

    from its case base is its main power. It build a structure that will return the most

    appropriate case(s) at high speed. Case base indexing minimizes the number of cases that

    have to be evaluated at run time and is required for a large set of cases as linear searched

    will yield a probability long retrieval time.

    Different methods:

    Nearest neighbor:

    Number of CBR system relates to nearest-neighbor method. The system

    would simply prefer cases that match more features to a case that matched fewer.The nearest-neighbor algorithm uses statistical method to determine the optimal

    set of feature and the number of case that should be used calculated similarities if

    the retrieval is somewhat flexible this approach works well. In Nearest-neighbor

    algorithm each new case is compared with all other cases in the database. As the

  • 8/7/2019 Applying AI in Insurance Domain

    14/29

    case base grows nearest neighbor cannot be calculated on the fly and pre-indexing

    is required.

    Induction:

    Inductive approaches to indexing are useful where the retrieval goal or

    case outcome is well defined. The output of the induction process is in the form of

    a decision tress. Induction-based system use a decision tree for retrieval as

    compared to nearest-neighbor indexing which is more associative, and induced

    decision tree is hierarchical and static.

    Advantage of induction method:

    o It can automatically, objectively, and rigorously analyze the cases to

    determine the best features for distinguishing them.

    o The resultant index increases the retrieval time by only the log of the

    number of cases rather than doing linearly. Retrieval time can be an

    important factor when using large case bases.

    Knowledge guided:

    A knowledge-guided approach uses human knowledge to the induction process by

    manually identifying known case features that are considered important and useful for

    case retrieval. Its the simplest approach to case classification and indexing. Cases are

    reviewed for their important features and the appropriate questions are passed to query

    the user about the existence or absence of features.

    [9] illustrates about the a hybrid Case-Based Reasoning (CBR) and

    Information Retrieval (IR) system that generates a query to the IR system by using

    information derived from CBR analysis of a problem situation. Based on a CBR analysis

    form a set of highly relevant cases the query is automatically formed by submitting in

    text. CBR is highly intelligent but limited in its reach and IR is broadly applicable but not

    able to reason in any depth. The goal in this project is to take advantage of the strengths

    of both CBR and IR in order to retrieve documents that are highly relevant to a problem

  • 8/7/2019 Applying AI in Insurance Domain

    15/29

    case from a standard IR collection without the need for creating symbolic case

    representations for documents in the collection. We address the issue of how

    automatically formulate good queries based on a problem situation in order to perform

    retrieval from large text.

    Hybrid CBR-IR system works by first performing a standard CBR

    analysis of the input problem case and then using the results of the CBR analysis to drive

    text-based document retrieval. Ordinarily, INQUERY would not engage in relevance

    feedback until retrieval. based on user input, had been made and a set of documents

    retrieved and presented to the user. In effect, our system uses feedback in the form of

    the RF-CKB on a null query. Our systems use of relevance feedback, in effect, tells the

    IR component that the cases found through the CBR analysis are highly relevant and that

    INQUERY should retrieve more like them. Note that while the CBR analysis is done

    with respect to the relatively small CKB available to the CBR component, and relevance

    feedback is done with respect to the even smaller set of special cases in the RF-CKB, the

    IR can be performed with respect to a text collection of arbitrary size. Instead of the user

    initiating the retrieval by making up a query, in our approach the user begins by inputting

    facts ot a case. In effect our system leverages its own m-house analysis of the problem

    case to a full-blown retrieval from an outside document base.

    [10] talks about the methods for case retrieval, reuse, solution testing, and

    learning are summarized, and their actual realization is discussed in the light of a few

    example systems that represent different CBR approaches. It is a problem solving

    paradigm that in many respects is fundamentally different from other major AI

    approaches. Instead of relying solely on general knowledge of a problem domain,

    or making associations along generalized relationships between problem descriptors

    and conclusions, CBR is able to utilize the specific knowledge of previously

    experienced, concrete problem situations (cases).

    Main types of CBR methods:

    Exemplar-based reasoning:

    CBR methods that address the learning of concept definitions are

    sometimes referred to as exemplar-based. The class of the most similar past case

    Fig 2

  • 8/7/2019 Applying AI in Insurance Domain

    16/29

    becomes the solution to the classification problem. The set of classes constitutes the

    set of possible solutions. Modification of a solution found is therefore outside the scope

    of this method.

    Instance-based reasoning:This is a specialization of exemplar-based reasoning into a

    highlysyntactic CBR-approach. To compensate for lack of guidance from general

    background knowledge, a relatively large number of instances are needed in order

    to close in on a concept definition.

    Memory-based reasoning: This approach emphasizes a collection of cases as a

    large memory, and reasoning as a process of accessing and searching in this

    memory. Memory organization and access is a focus of the case-based methods.

    1.3.3 Retrieval and Reuse

    Retrieve:

    Determine most similar case(s).

    Reuse:

    Solve the new problem re-using information and knowledge in the retrieved

    case(s).

    Revise:

    Evaluate the applicability of the proposed solution in the real-world.

    Retain:

    Update case base with new learned case for future problem solving.

    Fig [4] shows the CBR Cycle

  • 8/7/2019 Applying AI in Insurance Domain

    17/29

    Fig 5 consist of Task method decomposition of CBR. Tasks have node names in bold letters,

    while methods are written in italics. The links between task nodes (plain lines) are task

    decompositions, i.e part-of relations, where the direction of the relationship is

    downwards. The top-level task is problem solving and learning from experience and

    the method to accomplish the task is case-based reasoning.

    CBR Problem Areas:

    As for AI in general, there are no universal CBR methods suitable for

    every domain of application The challenge in CBR as elsewhere is to come up

    with methods that are suited for problem solving and learning in particular subject

    domains and for particular application environments. In line with the task model just

    Fig 4

  • 8/7/2019 Applying AI in Insurance Domain

    18/29

    Fig

    5

  • 8/7/2019 Applying AI in Insurance Domain

    19/29

    shown, core problems addressed by CBR research can be grouped into five areas. A

    set of coherent solutions to these problems constitutes a CBR method:

    Knowledge representation

    Retrieval methods

    Reuse methods

    Revise methods

    Retain methods

    [11] Present software architecture for CBR systems based on three

    components (a task description, a domain model, and adaptors) connected by a type of

    connectors called bridges. Adaptors are basic inference components that perform specific

    transformations to cases.The three main elements of the ABC software architecture are (i)

    a task description, (ii) a domain model, and (iii) a library of adaptors. These three

    elements connected with a special kind of connector called bridge. In addition, the

    problem to be solved is called input and for simplicity we will include the case base into

    the domain model element. The main issue to go from a specification like ABC to an

    actual implementation is deciding how is 1) the representation of components and

    bridges, and 2) The control scheme. We are implementing adaptors in Noos, a

    representation language designed for supporting knowledge modeling approaches to

    problem solving and learning in which different CBR systems have been built. In Noos

    cases are represented as feature terms, a formalism for representing structured cases in

    which any subpart of a case (feature term) is also a term and thus is also a case. Inference

    is provided by problem solving methods (PSMs) thast use domain knowledge to build

    models (or parts of models). A problem is solved when a case-specific model is

    completed, and then it is retained in the case base. Retrieval is performed by specialized

    PSMs, retrieval methods, that use domain knowledge or heuristic principles to search the

    case base. Concerning the control scheme, Noos inference is on demand, i.e. follows alazy evaluation strategy. The chain of control is thus backwards: retrieval methods

    determine the features of a case that they need, thus forcing the evaluation off.

    1.3.4 Issues in CBR:

    Retrieval and Reuse

  • 8/7/2019 Applying AI in Insurance Domain

    20/29

    Retrieve:

    Determine most similar case(s).

    Reuse:

    Solve the new problem re-using information and knowledge in the retrieved

    case(s).

    Knowledge Representation:A case is a unique knowledge entity describing a problem and a solution.A case

    can be represented as a single database object or broken into two or more associatedobjects. A typical case will have the following fields:

    Title

    Problem description

    Cause

    Solution

    Case Indexing:

    A CBR system its ability to retrieve relevant cases quickly and accurately

    from its case base is its main power. It build a structure that will return the most

    appropriate case(s) at high speed. Case base indexing minimizes the number of cases that

    have to be evaluated at run time and is required for a large set of cases as linear searched

    will yield a probability long retrieval time.

    1.4 Comparing ES and CBR:

    Expert system generates new knowledge and ES is to be replaced where as in

    Case Base Reasoning it searches for similar case and adapting these if necessary. In

    Expert system knowledge is stored implicitly while in CBR knowledge is stored

    explicitly. ES is hard to maintain as unpredictable implication by model change and

    extension. In CBR it is easier to maintain and update.

    2 Hybrid Intelligent Systems :

    2.1. Why hybrid systems?

  • 8/7/2019 Applying AI in Insurance Domain

    21/29

    There has been enormous interest in hybrid systems (especially neural-fuzzy,neural-genetic, and fuzzy-genetic) in the past ten years. Almost every conceivableproblem has been approached using some form of hybrid system. Why? Is thisbecause hybrid systems are universally bet terthan conventional approaches?

    One claim is that hybrid systems are intrinsically better. They allow for thesynergistic combination of two techniques with more strengths and less weaknesses

    than either technique alone.

    Although useful for many types of problem, hybrid systems provide even

    more opportunity for misuse than single techniques. Although motivated by

    combining the strengths of the system, the hybrid will, in the worst case, contain none

    of the strengths and all of the weaknesses of the component systems. While hybrid

    systems have great potential for solving some very difficult problems, they can also

    be used inappropriately. As a technique becomes more complex, the opportunities for

    misuse become greater, and hybrid systems are intrinsically more complex than single

    techniques. Many researchers are still making gross misuse of neural networks and

    fuzzy logic as single techniques, and you can expect that this will carry over into

    hybrid systems as they become more and more accessible.

    2.2. Integrating expert systems and case-based reasoning: approaches and

    applications:

    [12] Talks that this research involves both the development of intelligent

    systems and the study of cognitive models. The main motivations for the researches

    on HMs are

    Cognitive processes are not homogeneous, consequently, a large variety of

    representations and modeling techniques can be used

    The performance of intelligent systems can be improved by the

    combination of different Artificial Intelligence (AI) techniques. Therefore,

    may electively solve several real-world problems.

    [12] Presents new Case Based Reasoning approach using hybrid mechanisms for

    case retrieval and adaptation. Several strategies for case adaptation have been

    proposed in the literature. They can be classified in three main groups (see Fig. 6):

    substitutional

  • 8/7/2019 Applying AI in Insurance Domain

    22/29

    adaptation, transformational adaptation and generative adaptation.

    The strategies for substitution adaptation exchange solution attribute values of the

    retrieved solution by appropriate values, producing a new solution. The strategies for

    transformational adaptation modify the solution structure by including or removing

    components of the retrieved solution in order to satisfy the requirements of the new

    problem. The strategies for generative adaptation construct a new solution from problem

    data using a predefined procedure.

    The architecture of the proposed CBR system contains:

    A case retrieval mechanism composed by an ANN based on the

    Adaptative Resonance Theory (ART2) model;

    A case adaptation mechanism composed by one of the following ML

    algorithms:

    an ANN based on the Multi Layer Perceptron (MLP) model [4];

    a symbolic learning algorithm M5 [26];

    an algorithm based on the statistical learning theory named

    Support Vector Machine (SVM) [25].

    Case Retrieval and Incorporation Approach:

    The proposed strategy contains two levels of memory organization (see

    Fig. 2):

    o The first level is composed by the output layer of an ART2 network,

    which creates and also indicates clusters with similar cases, reducing the

    search space and the retrieval time.

    o The second consists of a simple flat memory that stores case instances

    grouped into the similarity clusters of the first memory level.

    Fig 6

  • 8/7/2019 Applying AI in Insurance Domain

    23/29

    Case Retrieval Process: The case retrieval mechanism proposed (first phase

    of the CBR CYCLE) allows the retrieval of one or more similar cases according to

    the system requirements. The memory organization employed allows the division

    of the search space, reducing the retrieval time.

    Case Incorporation Process: The case incorporation mechanism proposed supports the

    storage of new cases at any time (fourth phase of the CBR CYCLE). The memory

    organization used by this mechanism makes possible the storage of new cases without the

    eliminating cases previously stored.

    When a new problem is presented, an adapted solution is directly obtained by

    applying the procedure. The proposed memory organization and case retrieval

    mechanisms enable a search space reduction, decreasing the retrieval time required. The

    hybrid approaches would only search in the space corresponding to the cluster obtained

    by the ART2(Adaptive Resonance Theory 2) network. When CBR is applied to real-

    world problem solving, the retrieved solutions can not be directly used for a new

    problem. In general, they need to be adapted to new requirements. One of the major

    challenges in designing CBR systems is the acquisition and modeling of appropriate

    adaptation knowledge. The case adaptation approach proposed employs a process of

    adaptation pattern generation that can reduce the effort for knowledge acquisition in

    domains that require substitution adaptation. The hybrid system proposed is

    computationally cheap, since the generation of the adaptation patterns demands few

    solution components comparisons and the ART2 training for a pattern demands only one

    cycle. Moreover, the process to obtain an adaptation pattern data set is fully integrated

    with the case retrieval mechanism and can be implemented employing usual retrieval

    approaches. The results obtained suggest that the set of adaptation rules extracted from

    CB of the systems is consistent and that this approach of adaptation knowledge learning

    may be a potential technique for real-world problem solving.

    [13]Architecture was presented for improving system accuracy by bringing

    together knowledge in two forms: rules and cases. The architecture is intended for

  • 8/7/2019 Applying AI in Insurance Domain

    24/29

    domains that are understood well but not perfectly. The idea is that in such domains,

    expert knowledge in the form of rules can be used to provide a skeletal method for

    solving problems; cases are then used to flesh out the method by covering idiosyncrasies

    and special cases that were not anticipated by the rules. In addition to a reasonably

    accurate and efficient set of rules to serve as a starting point for problem solving, the

    architecture also needs knowledge in support of CBR-namely, a set of cases and a

    similarity metric. The set of cases should be extensive enough to illustrate the errors in

    the rules; any un-illustrated problems cannot be corrected. The architecture was applied

    to the task of name pronunciation. With minimal knowledge engineering, it was found to

    perform almost at the level of state-of-the-art commercial systems. More importantly, a

    modification experiment showed that its performance was higher than what it could have

    achieved with its rules or cases alone. This demonstrates the capacity of the architecture

    to improve upon a pure rule-based or case-based system. In addition to the accuracy

    benefits, having rules together with the cases allowed two innovations in CBR

    technology: first, the rules provided a natural way to index the cases (prediction-based

    indexing); and second, they provided a method of doing case adaptation, termed case

    adaptation by factoring. The architecture presented here is one data point in a hierarchy

    of possible hybrid approaches. One way to abstract away from its design is to keep the

    same reasoning components (RBR and CBR), but to combine them differently. The

    method of combination could be tailored to whatever knowledge is available in the

    domain, whether analytic (e.g., heuristics about when to believe RBR versus CBR) or

    empirical (e.g., examples of previous decisions combining RBR and CBR). Another way

    to abstract away from the architecture is to replace its RBR component with some other

    reasoning method.

    CBR then becomes a postprocessor to improve an approximate answer obtained

    by any method of choice. The downside, however, is that the benefits of having rules

    together with cases would be lost-alternative methods of case indexing and case

    adaptation would be needed. A final level of abstraction, and the one that is in fact the

    essence of the work presented here, is simply to combine multiple independent

    knowledge sources to achieve higher accuracy.

  • 8/7/2019 Applying AI in Insurance Domain

    25/29

    [14]This paper presents an approach to the retrieval of class diagrams integrating

    BN, CBR and WordNet. We describe how the BN is built from WordNet and from the

    case library, and a detailed example of network retrieval is given. One advantage of our

    approach, is the capability of assessing not only the similarity between diagram objects,

    but also the structure similarity of diagrams, through the use of network nodes

    representing diagram relations. This enables the BN to compute structural similarity,

    which is important for assessing diagram similarity. Another advantage is the leaning of

    new cases through the network updating. In relation, to other systems using BN for

    retrieval, our approach has the advantage of not depending entirely on cases for building

    the BN. An initial BN can be built using only WordNet and the query, which will then be

    updated with new cases.

    2.3 Tools for hybrid systems:

    There are number tools used to develop an hybrid expert system they

    are CLIPS (C Language Integrated Production System): A forward-chaining rule-based

    tool written in C by NASA. It can be easily embedded in other applications and includes

    an object-oriented language called COOL. It can be used on DOS, Windows, VMS, Mac,

    and UNIX. ART*Enterprise is the latest of the family of rule-based development

    environments originating with ART in the mid-1980s. It is a development environment

    for enterprise-wide applications, incorporating rules, a full object system which includes

    features currently not present in C++ or Smalltalk, and a large collection of object classes

    for UI development across platforms (from Windows to OS/2 to Unix), access to

    databases (SQL-based and ODBC-based), and multi-person development. The

    ART*Enterprise environment provides a forward chaining engine where backward

    chaining can be implemented, though it is not supported directly. ART*Enterprise also

    provides a CBR kernel for those who are interested in incorporating it into their

    applications.

    [2] CAPE combines the strengths of two well-established tools with

    very powerful but complementary pattern-matching mechanisms. Perl's ability to search

    text and match regular expressions is unequaled, while CLIPS provides powerful

  • 8/7/2019 Applying AI in Insurance Domain

    26/29

    mechanisms for finding patterns of combinations of symbolic information. The CAPE

    programmer can exploit the strengths of both, using Perl to analyze documents or query

    results, and CLIPS to recognize and react to the combinations of matches found.

    2.4 Issues in Hybrid Systems:

    There has been enormous interest in hybrid systems in the past ten years. Almost

    every conceivable problem has been approached using some form of hybrid system.

    Why? Is this because hybrid systems are universally better than conventional

    approaches?

    - Combining knowledge base and case representation:

    Here in Hybrid System combining CBR and knowledge representation is an

    most important task to be done and this task seams to be tedious job.

    - Overhead :

    Here indexing and searching of case are the overhead. This is to be stored

    with previous attributes.

    - Approaches :

    Different approaches are being used to over come this issues they are:

    o Sequence Infrence type

    o

    Knowledge conversational typeo Host assistant

    o Integrated reasoning type

    3 Intelligent Systems in Insurance Domain

    3.1. Potential applications of ES, CBR in Insurance:

    [15] Talks the underwriting function reviews applicant data for determination of

    insurability. The feasibility of utilizing an expert system in this area was examined and

    determined to be practical, and development of a prototype was initiated. The review

    of six expert system software packages narrowed the field to two viable candidates. The

    Ml package from technology was selected for the first prototyping stage. The prototype

  • 8/7/2019 Applying AI in Insurance Domain

    27/29

    was quickly developed with a very limited set of rules to provide a simple demonstration

    of the applicability of an expert system in this area. The host mainframe at the sponsoring

    insurance company proved incompatible with the Ml software, which was replaced by

    RuleMaster from Radian. The prototyping work was transferred to RuleMaster, which

    proved to be beneficial beyond the compatibility concerns. RuleMaster is a modular

    system as opposed to the inference network approach used in Ml, which more closely

    resembles LISP or PROLOG in structure. With the modular structure, knowledge and

    rule base building and modifications proved much easier from a programming standpoint

    3.2 Integrated approach applied in Insurance:

    The study of the insurance underwriting domain provided the basis of the

    prototyping and system structure for building the insurance underwriting expert system.

    The prototyping structure is based upon three key phases of the process: data entry, data

    revision and evaluation of data by the expert. The data entry phase was simplified by the

    use of the insurance industry standard ACORD application form to record and submit all

    applicant data. The data entry screens and data acceptance/storage routines were

    developed to portray the various fields from the ACORD forms, which facilitated entry of

    the data by a non-expert clerical person. All of the data required for the application

    evaluation and policy issuance is entered at one time. Data revision was incorporated into

    the data entry phase to allow for the correction of entry errors prior to acceptance of

    the data into the system for evaluation.

    During the evaluation phase the entered data items are processed according to the

    rules and structures in IU. The system determines a numerical weighted score for the

    application upon processing. Detail on the determination of the numerical scores, and

    verbal descriptions supporting the score assignments are displayed at the conclusion of

    the evaluation. UF&C also requested the inclusion of a conclusion strength factor, in a

    highly recognizable numeric form, as an additional supporting measurement of the

    conclusion reached. The overall control structure of IU is a forward chaining - data

    driven structure, particularly in the data entry and revision phases, with goal directed

    backward chaining structure used in the evaluation phase modules. The logical design of

    IU is based on the primary decision factors used in assessing the application. For

  • 8/7/2019 Applying AI in Insurance Domain

    28/29

    example, the automobile insurance underwriting portion of IU includes the review of

    information on the vehicles, drovers, agent submitting the application, and a standard set

    of ancillary information accompanying the application. The physical design follows the

    logical design through the modular structure facilitated by RuleMaster in the data entry,

    data revision and data evaluation phases of the process. Within each of these phases the

    structure is based on the primary decision factors listed previously.

    4 Future study

    4.1. Studying the potential applications where hybrid approach can be used

    Studying the potential applications of different hybrid approach which we

    are going to use.

    4.2. Devising methodology for knowledge acquisition, presentation and retrieval.

    We are study the overall different methodology for knowledge acquisition,

    presentation and retrieval. Using this we are going to representing it to our Insurance

    domain.

    4.3. Selection/customization of proper tool

    There are different tools that we have to study. The following are some of

    proper tool:

    - LISP

    - PROLOG

    - JESS

    - IKEN Core

    4.4. Developing prototype system/s

    Here we are going to implement prototype systems for our insurance domain.

  • 8/7/2019 Applying AI in Insurance Domain

    29/29

    Reference: