Tıpta uygulamalar için bir karatahta modeli

Abstract

Uzman sistemler yetmişli yıllardan başlayarak yapay zekanın en önemli araştırma alanlarından biri olmuşlardır. İnsan davranışlarının bir simülasyonu olarak çıkarım yapabilen bu sistemler geniş bir problem alanının özel bir bölümünde, tam bilgiyle donatılarak problemleri çözebilirler. Bilginin her zaman kesin olmaması sorunu, önceleri Bayes olasılık methoduyla aşılmaya çalışılmış daha sonraları, ilk olarak MYCIN bilgi mühendisleri buna itiraz ederek 1950 yıllarına dayanan Carnap teorisinden hareketle, ikinci bir olasılık yaklaşımı olan belirlilik çarpanlarını kullanmışlardır. Farklı belirlilik çarpanlarına sahip bilgi tabanları olan iki uzmanın, aynı belirtilere getirdiği değişik yüzdeli çözümler, basit bir karatahta modeliyle bir kontrol ünitesinde birleştirilerek tek bir sonuca indirgenmiştir. Uzman sistem, girilen sonuçtan hareket ederek bu sonucu sağlayan şartlan sorgulayan geriye doğru çıkarım mekanizması kullanmaktadır. Tezin birinci bölümünde problemin genel tanıtımı yapılırken, ikinci bölümde uzman sistemler ve çıkarım mekanizmaları anlatılmıştır. Üçüncü bölümde belirlilik çarpanları ve karatahta modeli anlatılmıştır. Dördüncü bölümde bilgi tabanının oluşturulması, kullanılan veri yapılan, genel bilgi tabanı ve programın çalıştırılması anlatılıp örnekler verilmiştir.

In this thesis, we construct a blackboard model to manage the uncertainties involved in medical diagnosis. We have considered the diagnosis of two doctors of children diseases. Furthermore we have assumed that the doctors have the same educational background, but they have gained experiance in different environments. So that our experts who cooperate on the blackboard have the same knowledge base, yet the rules have different certainty factors. Since MYCIN'S knowledge engeneers have found succesfiill applications in practice. We have modeled uncertainty in diagnosis with certainty factors. The expert system has backward chaining inference and LIFO (last in first out) stack structure. The system asks for a conclusion from the user and search for the rules which give for the conclusion. If the system finds a rule, which includes same conclusion, asks the conditions that provide the conclusion. Every conditions and conclusions have certainty factors that derived from a human expert. Blackboard control unit combines two certainty factors into one certainty factors and write to the blackboard. Control unit takes a conclusion, passes it two experts and waits for the rule with different certainty factors to be activated. In blackboard model, expert system has a knowledge base but it is prepared outside of the computer program. This unique characteristic is superiority of this model to other expert system models. Our system has two knowledge bases that might write in any tex editor by a knowledge engineer. Control unit manages the chainging on blackboard, decides which rule to be activated, keeps the values of conditions in working memory and continues until to come to a conclusion. Our two expert systems have a rule based knowledge base. The rules are written " If-Then " type. Every rule might have twenty conditions that depends on " and ", " or " operators. The knowledge base has fifty five rules. Today, expert systems have thousands rules to be more powerful. A human expert of children diseases may easily increase or update our knowledge base thus, expert system may be more efficiency. Expert systems becomee very important area of artificial intelligence since seventies. "By definition, an expert system is a computer program that simulates the thought process of a human expert to solve complex decision problems in specific domain of wide area. An expert system operates as an interactive system that responds to questions, asks for clarification, makes recommendation and generally aids the decision making pro-cess" [4]. Knowledge is not in certainty every time in life. This problem firstly were solved Bayesian Probability, after several years MYCIN knowledge engineers disagreed this method and they used certainty factors derived Carnap's theory in 1950 [1]. Carnap distinguished two types of probability. One type of probability is ordinary probability associated with the frequency of reproduciple events. The second type is called epistemic probability or the degree of confirmation, because it confirms a hypothesis based on some evidence. This second type is another example of the degree of likelihood of a belief. In MYCIN, the degree of confirmation was originally defined as the certainty factor, which is the difference between belief and disbelief. CF(H, E) = MB(H, E) - MD(H, E) Where number. CF is the certainty factor in the hypothesis H due to evidence E MB is the measure of belief in H due to E MD is the measure of disbelief in H due to E The certainty factor is a way of combining belief and disbelief into a single The measures of belief and disbelief were defined in terms of probabilities by MB(H,E) = P(H) = 1 otherw ise maxfP(HlE), P(H)]- P{H) max[l,0]- P{H) MD(H,E) = « P(H) = 0 otherw ise min[P(H|E), P(H)"|- P(H) min[l,0]- P(H) According to these definitions, some characteristics are below. vi The original definition of CF was CF = MB - MD There were difficulties with this definition because one piece of disconfirrning evidence could control the confirmation of many other pieces of evidence. For example, ten pieces of evidence might produce a MB = 0.999 and one confirming piece with MD = 0.799 could then give CF = 0.999 - 0.799 = 0.200 In MYCIN, a rule's antecedent CF must be > 0.2 for the antecedent to be considered true and activate the rule. This threshold value of 0.2 was not done as a fundemantal axiom of CF theory. Thise definition greatly reduce the system's efficency. The definition of CF was changed in MYCIN in 1977 to be CF = MB- M D \~mm(MB,MD) to soften the effects of a single piece of disconfirrning evidence on many confirming pieces of evidence. Under this definitions vn CF = 0.999 - 0.799 0.200 1 - min(0.999,0.799) 1 - 0.799 = 0.995 which is very different from previous definition. Now the 0.995 will cause the rule to be activated. MYCIN methods for combining evidence are below. £, and E2 Ex or E2 not E mm[CF(H,Ex),CF(H,E2)] max [CF(#,£,), CF(H,E2)] - CF{H,E) The fundemantal formula for the CF of a rule lFe THEN H is given by the formula CF(H,e) - CF(E,e) CF(H,E) CF(E,q) is the certainty factor of the evidence E on uncertain evidence e. CF(HE) is the certainty factor of the hypothesis. CF(H,e) is the certainty factor of the hypothesis based on uncertain evidence e. Suppose another rule also concludes the same hypothesis, but with a different certainty factor. The certainty factors of two rules concluding the same hypothesis is calculated from the combining function for certainty factors defined as CF (CF CF) = w cemimi 1>W 2/ CF, +CF2(\-CF1) CF1+CF2 1-mindC^UCFJ) [CF^CFfi+CFJ both>0 one <0. both<0 where the formula for CF, COMBINE used depends on whether the individual certainty factors are positive or negative. vin Values of CFCOMmm range between -1 and +1. A CF of a negative number indicates a predominance of opposing evidence for the rule being correct. A positive CF indicates a predominance of confirming evidence for the rule being correct. Therefore, a CF of +1 indicates absolute certainty that the rule is correct, while a CF of -1 indicates absolute certainty that the rule is incorrect. A group of multiple inferences that connect a problem with its solution is called a chain. A chain that is searched from a problem to its solution is called a forward chain. Another way of describing forward chaining is reasoning from facts to the conclusions which follow from the facts. A chain that is searched from a hypothesis back to the facts which support the hypothesis is a backward chain. Another way of describing a backward chain is in terms of a goal which can be accomplished by satisfying subgoals.