Case-based Modeling
Otakar Babka, Vong Chi Man
University of Macau, Faculty of Science and Technology
P.O.Box 3001, Macau (via Hong Kong).
email: ,
Keywords: Knowledge-based systems, Case-based reasoning, Rule-based reasoning, Soil settlement, Reclamation areas.
ABSTRACT
Case-based reasoning, partially augmented by rule-based reasoning, is studied in the paper from perspective of modeling. Applied to reclamation areas, this combined technology has been adopted to the support of soil settlement modeling. Case-based reasoning as a principal paradigm of the presented concept, assists to selection of the best mathematical model for modeling, as well as to estimation of the ultimate settlement. Modeling framework has been built as a part of a planned Geographical Information System monitoring and predicting reclamation area parameters.
INTRODUCTION
Case-based reasoning paradigm, supporting modeling and consequent decision making processes, is studied in the paper. In some situations, the framework can also be partially assisted by rule-based support if appropriate. This combined technology has been adapted to reclamation area, monitoring and predicting settlement of soil. Modeling of the settlement process is an inevitable part of this effort and the supporting framework can contribute substantially to smooth the modeling. Case-based reasoning was adopted as the key paradigm to the selection of the best mathematical model, as well as to a direct prediction of the ultimate settlement.
The next paragraph of the paper focuses principles of case-based reasoning. Reclamation area settlement modeling is described in the third paragraph.
Case-based reasoning
Case-based reasoning has been adopted as a key paradigm of the framework supporting modeling in decision support system. This system has been applied to reclamation area settlements as a part of a wider geographical information system.
Case-based reasoning is one of the most effective paradigms of knowledge-based systems. This promising concept addresses successfully main problems of the traditional knowledge-based systems. Despite great success of them in the last decades, these traditional knowledge-based systems face several serious problems. Watson and Marir [5] outlined the main issues of the traditional paradigms as the following:
(1) Knowledge elicitation bottleneck (knowledge elicitation is a difficult and tedious process).
(2) Implementing a knowledge-based system, using the traditional paradigm is a lengthy process requiring cooperation of high qualified experts.
(3) Traditional paradigms are mostly unable to manage effectively large volumes of knowledge.
(4) It is difficult to maintain a system based on a traditional paradigm during its life cycle.
The above problems can be partially solved by case-based reasoning. Differing from more traditional methods, case-based reasoning is one of the most effective paradigms of knowledge-based systems. Relying on case history is its principal feature. For a new problem, case-based reasoning strives for a similar old solution. This old solution is chosen based on a correspondence of the new problem to some old problem (which was successfully solved by this solution). From this perspective, case-based reasoning represents a strongly anthropomorphic approach. Solving a new problem, people usually center on some old solution that was successful for a similar problem in the past. So, previous significant cases are gathered and saved in a case library.
A case-based reasoning system can only be as good as its case library [4]. This implies two important issues. Only successful and sensibly selected old cases should be stored in the case library. Each candidate is chosen after an evaluation only.
Representation and indexing are other important issues of this methodology. The description of the case should comprise the problem, solution of the problem, and any other information describing the context for which the solution can be reused. A feature-oriented approach is usually used for the case description. A sensible identification of important features is one of several crucial decisions.
The case library serves as a knowledge base of the case-based reasoning system. The system can learn by acquiring knowledge from these old cases. Learning is basically achieved in two ways:
(i) accumulating new cases, and
(ii) through the assignment of indexes.
The first possibility basically means a quantitative improvement, while the second one can improve the case library rather qualitatively.
When solving a new case, the most similar old case is retrieved from the case library. The suggested solution of the new case is generated in conformity with this retrieved old case. The search for a similar old case from the case library represents an important operation of the case-based reasoning paradigm. Retrieval relies basically on two methods: nearest neighbor, and induction. Complexity of the first method is linear, while the complexity of the search in a decision tree, developed by the latter method, is only logarithmic. This reasonable complexity enables the effective maintenance of large case libraries.
The case-based reasoning approach prefers rather verified ways, in accordance with the anthropomorphic and pragmatic approach. Why to start from scratch again if we can learn from the past? Once a suitable old solution, tied to the new context, can be retrieved by the case-based reasoning part of the framework, such a tested solution should be preferred.
Case-based reasoning relies on the idea that situations are often repeating during the life cycle of an applied system. This is usually a realistic assumption. Further, after a short period of time the most frequent situations should be identified (and documented in the case library). So, the case library can usually cover common situations after a short time. However, when relying on case-based reasoning exclusively, two problems can be encountered: (i) it is difficult to start from the very beginning with an empty case library, (ii) on the other hand, after some time the case library can become huge and contains much redundancy. Therefore it is better to combine the case-base reasoning with some other paradigm to compensate for these marginal insufficiencies. The rule-based reasoning component can cover these gaps with the help of implemented knowledge and heuristics. This means that the rule-based part is capable of suggesting its own solution. This suggestion will be requested especially under the following circumstances:
(1) No suitable old solution can be found for a current situation in the case library. For this event, the rule-based reasoning part can be activated. However, once generated by this part of the framework, such a solution is then evaluated and tested more carefully.
(2) Some situations may be almost the same but not identical. Such cases cause a high level of redundancy of the case library. To replace such a class of very similar cases by a set of rules can partially solve this problem.
(3) The rule-based reasoning part can also cooperate in solving some specific situations. So, in an effort to extend its capability, the rule-based paradigm is augmented [1,3]. Along with knowledge in the form of rules, programs can also be integrated into the framework. As a built-in tool, a selective mechanism is implemented and encapsulated in the framework. In terms of the control of a computing process, the mechanism can facilitate two basic operations: (i) selection of the most suitable source (i.e. a piece of knowledge, models, or some other programs) integrated into the framework, and (ii) sequencing of the selected programs, according to a given global goal and input data.
The selective mechanism can either control the computing process autonomously or, the user can be substantially supported on the above points. Software systems for modeling are frequently large. It appears to be difficult to utilize such complex systems efficiently without intelligent support for the user. Such support can be provided on two levels: (i) The user can be guided and advised during the process of modeling. (ii) An evaluation of the results and a consequent decision how to utilize these results can be supported by the framework. The user’s decision making process is facilitated by utilizing incorporated expert knowledge and other related sources of information maintained by the framework.
Knowledge bases consist of three types of knowledge: (1) Several general heuristics can contribute to the optimal solution search of a very wide class of tasks. (2) However, the dominant part of the knowledge support is based on domain-specific knowledge. (3) For a higher efficiency, metaknowledge is also attached to the knowledge base. This “knowledge about knowledge”, in the form of metarules, can contribute considerably to a smooth reasoning process.
Settlement Modeling
Considering the above discussion, case-based reasoning was implemented for our decision support system, along with rule-based reasoning. The first paradigm proved especially suitable to this kind of application, reclamation areas issue.
Reclamation areas are important and costly zones. Monitoring and prediction of soil settlements of the reclamation areas is an inevitable and crucial source of information about the area. As a part of a wider geographical information system, the decision support system has been developed and applied to monitoring of settlements. Its main tasks are:
(1) settlement process monitoring and modeling,
(2) final settlement prediction,
(3) hazardous states recognition,
(4) recommended intervention.
The cases are described with the help of the chosen important features. The sensitive choice of these features is a crucial decision, predestining the character and capability of decision supporting systems using this reasoning paradigm. In the presented application, each settlement point is assigned to a case. Geographical and geotechnical data of this point and the process of the settlement describe the case.
In agreement with the purpose of our system, features describing the geographical and geotechnical nature of a point mean input features and are known for all examined points. The inquired data about these points are called output features. In the presented application, features describing the process of the settlements and recommended activities for hazardous situations represent the output features.
For previously examined cases, both input features and output features are known and recorded. So, these old cases can serve as a source of knowledge of case-based reasoning system. Learning process of case-based reasoning system passes through the accumulation of the old cases. Based on this knowledge, output features can be consequently found for similar newly examined cases.
For the presented application, recorded points of earlier completed reclamation areas can serve for this purpose. Features of selected points are collected in a case library. Once collected, case-based reasoning can predict the process of settlement for examined points of a new reclamation area, if their geotechnical situation is similar. Examining a point of the new area, the case corresponding to a previously examined point with the most similar geotechnical features is retrieved from the case library. To fit closer to the new situation, the retrieved old case is then adapted. Taking into account the differences between features of the new and the old retrieved point, output features are recalculated before suggesting new results.
Besides detection of a hazardous situation, the main concern of a user of our system is prediction of the settlement process and the final value of this settlement. Consulting the system, a suggestion for the user can be generated by two alternative levels of the system:
· Settlement process and its final value for a new situation are suggested according to settlement of the retrieved similar previous case, adapting the old results in conformity with the new situation.
· The process of settlement is traditionally estimated with the help of various mathematical models, each corresponding to a different class of situations. For a user it is difficult to choose the best model for a given situation. However, the most suitable model can be recommended by the decision support system. The model is chosen among several built-in models. Using this chosen model, the process of settlement is then calculated, including an estimation of its final value.
AKNowledgment
Ms. Lorinda Aguiar Gomes Garanito participated in the research, especially in the application to reclamation areas.
References
[1] O. Babka, Knowledge-based System Supporting Design System. European Journal of Engineering Education, Vol. 17, pp. 181-187 (1992).
[2] E. Turban, Decision Support and Expert Systems: Fourth Edition, Prentice Hall International, Inc., London (1995).
[3] P. Harmon, C. Hall, Intelligent Software Systems Development, John Wiley (1993).
[4] J. Kolodner, Case-based Reasoning, Morgan Kaufman Publ., San Mateo, CA, U.S.A. (1993).
[5] I. Watson, F. Marir, Case-based Reasoning: A Review. The Knowledge Engineering Review, Vol. 9, No. 4 (1994).
Case-based Modeling
Otakar Babka, Vong Chi Man
University of Macau, Faculty of Science and Technology
P.O.Box 3001, Macau (via Hong Kong).
email: ,
Keywords: Knowledge-based systems, Case-based reasoning, Rule-based reasoning, Soil settlement, Reclamation areas.
ABSTRACT
Case-based reasoning, partially augmented by rule-based reasoning, is studied in the paper from perspective of modeling. Applied to reclamation areas, this combined technology has been adopted to the support of soil settlement modeling. Case-based reasoning as a principal paradigm of the presented concept, assists to selection of the best mathematical model for modeling, as well as to estimation of the ultimate settlement. Modeling framework has been built as a part of a planned Geographical Information System monitoring and predicting reclamation area parameters.
References
[1] O. Babka, Knowledge-based System Supporting Design System. European Journal of Engineering Education, Vol. 17, pp. 181-187 (1992).
[2] E. Turban, Decision Support and Expert Systems: Fourth Edition, Prentice Hall International, Inc., London (1995).
[3] P. Harmon, C. Hall, Intelligent Software Systems Development, John Wiley (1993).
[4] J. Kolodner, Case-based Reasoning, Morgan Kaufman Publ., San Mateo, CA, U.S.A. (1993).
[5] I. Watson, F. Marir, Case-based Reasoning: A Review. The Knowledge Engineering Review, Vol. 9, No. 4 (1994).
GMD FIRST
Congress Office IMCS WC’97
Rudower Chausse 5
D-12489 Berlin
Germany
Subj.: Author’s Collective Extension
April 24, 1997
Enclosed I am sending our papers:
· O. Babka, Vong Chi Man, Case-based Modeling
· Vong Chi Man, O. Babka, Knowledge-based Support of Simulation
for the conference IMACS WC’97, Berlin, August 24-29, 1997.
In the last stage, Mr. Vong Chi Man contributed to the described research very substantially. So, I needed to invite him as an co-author of the both papers. I would like ask you to cite his name in all conference information booklets and materials.