251 An Approach To Develop Knowledge Modeling Framework Case Study Using Ayurvedic Medicine D.S. Kalana Mendis *, Asoka S. Karunananda +, and U. Samarathunga ++ * Department of Information Technology, Advanced Technological Institute, Labuduwa, Sri Lanka Email: kalanaatil@mail.com + Faculty of Information Technology, University of Moratuwa, Sri Lanka Email: Asoka@itfac.mrt.ac.lk ++ U. Samarathunga Gampaha Wickramarachi Ayurveda Institute, University of Kelaniya, Sri Lanka Abstract This paper provides a methodology aimed at better measuring tacit knowledge in Ayurvedic medicine. Among other areas, western medicine has heavily used the expert systems technology. There are very little applications of expert system technology for alternative systems of medicine. Since these areas have more implicit form of knowledge, the development of expert systems for such areas contributes both expert system technology and domain of alternative systems of medicine. This paper describes our research into development of expert systems for Ayurvedic medicine, which has won a big recognition on these days. Since Ayurvedic medicine is strongly based on the concept of individuality for diagnosis of diseases and prescription of treatments, the research has been conducted to model Ayurvedic knowledge on recognition of individuals. An expert system has been developed to recognize human constitution according to Ayurvedic medicine. A statistical technique of principle component has been used to analyze the classification of individuals. Fuzzy-expert system has been developed to model knowledge by eliminating the limitations associated with the PC analysis. The system has been tested against the medical practioners of Ayurvedic medicine and gained 77% accuracy. It is found that the fuzzyexpert system developed can be used as a decision support system tool for Ayurvedic practioners and as a learning system for Ayurvedic medical students. The system has also been developed as a framework that can be used to model Ayurvedic like domains where the knowledge is rather implicit or tacit. So, PC and Fuzzy logic has been developed as modules that can be integrated with a standard expert system shell. I. INTRODUCTION Ayurvedic medicine has proposed a different approach to western medicine for diagnosing of diseases and prescription of treatments. In particular Ayurvedic medicine takes infidelity of patients into consideration. Stated in another way, treatments are based on individual differences. In this sense, Ayurvedic medicine also provides set of knowledge for classification individuals. This knowledge has been presented in the form of questionnaire and used for many years without much research into its development. According to medical practitioners of Ayurveda, it appears that the questionnaire has repetitive and inconsistent information. This paper researches into development of a knowledge-modeling framework, which emulates Ayurvedic knowledge of classification of individuals as a case study. As such we decided to resolve the problem with the help of Artificial Intelligent techniques (AI). It is well known fact that AI techniques are better at solving real world problems, which cannot be solved otherwise. In particular AI techniques can be used to models domains with less formal knowledge. Among other AI techniques, we have used Fuzzy logic to fine tune the results obtained from PC analysis. Finally the system has been developed as an Expert System, which models Ayurvedic classification of individuals. With this technology the system has added features such as incorporating new knowledge, explaining reasons for answers given. So, the expert systems for individual classification are not only a tool for diagnosis but also a learning system for Ayurvedic medical students. II. AYURVEDIC CLASSIFICATION OF INDIVIDUALS Ayurvedic medicine has a very strong bearing on the concept of Prakruti, which means nature (natural form) of the build and constitution of the human body. According to Ayurveda the path to optimal health is different for people depending on their Prakruti. For individuals the Prakurthi is defined as a combination of (Vatha, Pittha and Kapha). A balanced state of the Tridoshas makes a healthy and balanced person (Physically and Mentally) [6]. Since we all have different combinations of the Tridoshas, The diagnosis of pakruti offers unique insights into understanding and assessing one s health. It is not merely a diagnostic device but also a guide to action for good health. It assesses the, dominance of Tridoshas and gives advice for preventive and
252 primitive health care. The ancient science of Ayurveda is the oldest known form of health care in the world. Recognition of human constituent in Ayurveda, is currently based on a standard questionnaire on subjective criteria based on ancient theories of Ayurvedic scholar Charaka, 1000 BC and Susruta, 600 BC. Questions in concerned are very much user-friendly and based on medical theories of Ayurveda, which is used for finding constituent type, has probes such as repeating questions and classification of constituent type. This has been used for classification of individuals for many centuries. There has been no research into improve the questionnaire although people have realized that the classification is not acceptable sometimes. III. FRAMEWORK FOR MODELING KNOWLEDGE The approach has been converted for an implementation using the architecture [10] given below (Figure 1). It is consisted of with modules such as principle component analyzer, database, knowledge base, and fuzzy logic module and inference engine. Fig. 1. Top-level architecture A. Principle Components Analyzer Tacit knowledge [12] has been extracted from the expert and formulated in a questionnaire. It is evaluated using Likert scale technology. In the first instance of knowledge acquisition, a pilot survey has been done for the purpose of extracting principal components [7]. The SPSS [18] is used for conducting the functions of principle components extracting. B. Fuzzy Logic Module The output results of the principle component analyzer would be the input for the fuzzy logic [4] module. In the case of generating membership function, finding the interval is considered as an automated process in this module due to instead of using runtime inputs. This module has been implemented using Visual Basic for widening scope of generating membership function. Further, fuzzy rules have been constructed in the fuzzy logic module. C. Database Extracted principle components have been stored in Ms Access database, which integrated with the principle component analyzer through the developer interface that is considered as a sub interface of the user interface. The questionnaire consisted of tacit knowledge also been stored in the database that integrated with the user interface. D. Knowledge Base Explanations for output generated by the fuzzy logic module have been processed using fuzzy rules in the knowledge base. Further, knowledge engineer is given a facility to add new rules in the runtime. The knowledge base has been implemented using FLEX [9] expert system shell, which embedded in WinProlog E. User Interface The user interface facilitates for both developer and general user. Once knowledge engineer develops a particular framework for required tacit domain with interaction of the expert, and then general user will be given a facility of using the framework for decision-making purposes. So, it has been divided the user interface in developer interface and general user interface. General user will be able to use a developed framework using a questionnaire, which has been implemented as a web page linked to the database. F. Inference Engine The inference engine carries out the reasoning whereby the expert system reaches a solution. This is the inference engine of the FLEX expert system shell. Since this is built in to the system there is no development activities with regard to this component in the system. Note that inference engine has nothing to do with the modeling of tacit knowledge but it runs the expert system. IV. FRAMEWORK IN PRACTICE In the exciting system, the method of analyzing constituents is not consistent. Although Ayurvedic practitioners use a questionnaire but leads several problems like dependencies among the questions in the questionnaire and analysis of the constituent type. We addressed these problems to solve using following stages [19]. A. Removing dependencies It is consisted of 72 questions to analyze vata, pita and kapha. We have done a pilot survey for 100 no. of students for statistical modeling using the questionnaire. Principle component analyzer has been used to remove dependencies in the questionnaire. It has been identified 25 principal components using SPSS as shown in matrix given below. Here V1, V2..V24, K1, k2..k24, P1, P2..P24 denotes question-numbering system in the questionnaire.
253 1 2.. 24 25 V1-0.228622 0.249362. -0.073945 0.058179 V2 0.08431 0.20654. -0.097192-0.112795. V=. V23-0.645803 0.232312. 0.0067-0.083959 V24-0.222147-0.06453. -0.073514 0.084404 K1 0.012511-0.096332. 0.141314 0.25113 K2-0.005642 0.268145. -0.179992 0.111715. K= M =. K23 0.409442 0.073812. -0.115118-0.056431 K24 0.696973 0.126679. 0.098213 0.045471 P1 0.430044 0.14608. 0.023669 0.09045 P2 0.243781 0.373485. -0.040468 0.149644.. P= P23 0.009727 0.012529. -0.072224 0.177827 P24-0.378091 0.096985. 0.158006 0.069821 0 X=<X L V (X) = (X-X L )/(X U -X L ) X L <X<X U 1 X=>X U V(x) denotes membership function for classifying vata constitution. This has been constructed using Visual Basic. Co-matrix computed in principle component analysis is given below. V= -0.228622 0.249362. -0.073945 0.058179 0.08431 0.20654. -0.097192-0.112795-0.645803 0.232312. 0.0067-0.083959-0.222147-0.06453. -0.073514 0.084404 24*25 B. Analysis of human constituents Human constituents can be computed in to vata, pita and kapha in percentages as shown below. Membership functions for vata, pita and kapha have been constructed in fuzzy logic module using the out puts of principle component analyzer as illustrated in Fig.2. C. Fuzzy logic for Analysis of human constituents Human constituents can be computed in to vata, pita and kapha in percentages as shown below. Membership functions for vata, pita and kapha have been constructed using fuzzy logic based on out puts of principle component analysis. - Membership function for classifying Vata constitution Boundary values of membership function have been constructed using the output of the principle component analysis. Fig. 2. Analysis window D. Explanations for derived human constituents Explanations for output generated by the fuzzy logic module have been processed using fuzzy rules in the knowledge base in the expert system. The knowledge base has been implemented using FLEX expert system shell, which embedded in WinProlog. In relation to Ayurvedic domain, possible diseases can be occurred due to dominated constituent type. It is illustrated as shown in Fig. 3. Q X L Q X U = 1 = 6 25 24 i= 1 j= 1 25 24 i= 1 j= 1 a = 8.510004 (1) ji a = 51.06002 (2) ji Here X L denotes lower bound value at the minimum level of evaluation scale (Does not apply) in the questionnaire. X U denotes upper bound value at the maximum level of evaluation scale (Applies most) in the questionnaire.
254 id Vata pitta kapha Expert_decision 70 30.28 29.58 40.14 KV 71 12.71 44.92 42.37 PK 72 11.18 40 48.82 PK 73 11.24 40.24 48.52 PK 74 23.44 26.9 49.66 PK 93 17.09 36.75 46.15 KV 94 33.09 30.15 36.76 KV The evaluation was conducted to see far the answers generated by the system matches with the identification by Ayurvedic experts and the students. Further, the system s ability to fine-tune the answers was also tested. It has been investigated that 23 (77%) of conclusions matches with the system and expert (see Table 2), which leads to determine the accuracy of the system. Fig. 3. Explanations window V. TESTING The expert system developed using this approach was tested with a group of 30 persons of Ayurvedic experts and students (see Table 1). TABLE 1 SYSTEM TESTING: EXPERT VS. SYSTEM id Vata pitta kapha Expert_decision 14 25.71 20.71 53.57 KV 15 32.95 23.86 43.18 VP 16 39.88 23.81 36.31 VP 17 27.65 46.1 26.24 KP 22 25.69 29.36 44.95 KV 23 33.58 24.09 42.34 KV 24 25.71 34.28 40 KP 26 32.21 31.54 36.24 KV 27 22.51 29.8 47.68 KP 28 20.37 30.56 49.07 PK 35 30.6 35.52 33.88 PK 46 29.71 17.39 52.9 KV 47 41.07 10.71 48.21 KV 48 34.5 32.16 33.33 KV 49 23.46 28.57 47.96 PK 50 35.27 30.77 33.97 KV 52 42.36 36.11 21.53 VP 56 23.01 35.71 41.27 PK 65 47.94 19.86 32.19 KV 66 14.03 35.96 50 PK 67 19.15 36.88 43.97 PK 68 22.46 25.36 52.17 PK 69 40.47 26.78 32.74 PK TABLE 2 COMPARISSION OF CONCLUSIONS: EXPERT VS. SYSTEM id vata pitta kapha Expert_decision conclusion 14 25.71 20.71 53.57 KV matched 23 33.58 24.09 42.34 KV matched 24 25.71 34.28 40 KP matched 26 32.21 31.54 36.24 KV matched 27 22.51 29.8 47.68 KP matched 28 20.37 30.56 49.07 PK matched 35 30.6 35.52 33.88 PK matched 46 29.71 17.39 52.9 KV matched 47 41.07 10.71 48.21 KV matched 48 34.5 32.16 33.33 KV matched 49 23.46 28.57 47.96 PK matched 50 35.27 30.77 33.97 KV matched 56 23.01 35.71 41.27 PK matched 65 47.94 19.86 32.19 KV matched 66 14.03 35.96 50 PK matched 67 19.15 36.88 43.97 PK matched 68 22.46 25.36 52.17 PK matched 70 30.28 29.58 40.14 KV matched 71 12.71 44.92 42.37 PK matched 72 11.18 40 48.82 PK matched 73 11.24 40.24 48.52 PK matched 74 23.44 26.9 49.66 PK matched 94 33.09 30.15 36.76 KV matched The system facilitated to derive constituents types in percentages while Ayurvedic experts obtain only the constituent type. As recommendation given by the Ayurvedic experts, determining constituent s types in percentages is an important criterion for prescribing drugs for a disease. Further, our system provide as an option to
255 find out possible diseases. In generally, the system can be used as a self-assessment for finding constituents. According to Ayurvedic medicine, regiments can be done easily by knowing the constituent type. The human constituents can be computed as a combination. So it would help to find the effectiveness of minimum type in a diagnosis. VI. CONCLUSION The knowledge-modeling framework developed can be used as a tool for supporting Ayurvedic medical practitioners for recognition of human constituents. Even an ordinary person can also use this without consulting an Ayurvedic practitioner. Further, Ayurvedic medical students can use the system as a learning system. The users of the system are not expected to hold knowledge in statistical or artificial intelligence techniques. This system can also maintain history of patients for research related human constitutes. It should be noted that with the help of Artificial Intelligence technologies we have improved the correctness of the decision making process in relation to the use of traditional questionnaire. This eliminates the inconsistencies and repetitiveness of answers and also provides a means for explanation of reasons for answers. The system can be further developed as a comprehensive framework with access to several domains with tacit knowledge. [12] D. Richards and P. Bush, Measuring, Formalizing and Modeling Tacit Knowledge IEEE/Web Intelligence Conference (WI-2003) Bejing. [13] R. Dawson, (2001), Developing Knowledge-Based Client Relationship, Butterworth Heinemann. [14] S.N. Tripathi (1978), Clinical Diagnosis, Science and Philosophy of Indian medicine. [15] XpertRule Knowledge Builder, www.attar.com. [16] V. Noak (2000), Discovering the world with Fuzzy logic, A Springer Verlag Company, PP. 3 50. [17] http://kmi.open.ac.uk/knowledge-modelling/people.html [18] M. C. Morogan, (1997), SPSS for windows, release 8.0.0, DSV KTH SU, Sweden [19] D.S.K. Mendis, A.S. Karunananda and U. Samarathunga (2005), An intelligent system for smart thinking in clinical diagnosis, International Information Technology conference, Colombo, Accepted for publication. REFERENCES [1] K-P. Adlassnig (1986), Fuzzy set theory in medical diagnosis, IEEE Tr. On Syst., Man, and Cybernetics 16(2), March/April, 260-265. [2] R.D. Zielstorff (1998). Online practice guidelines, JAMIA 5, 227-236. [3] W. Jim, B. Gleb, V.Z. Berend (2000) Fuzzy logic in clinical practice decision support systems, Proceedings of the 33rd Hawaii IEEE International Conference on System Sciences pp. 1-10 [4] J.K.George, B.Yuan(1995), Fuzzy sets and Fuzzy logic, prentice hall of India, pp. 280 300. [5] L. Jonson (1988), Expert system Architectures, Kopan Page Limited [6] G.P. Dubey (1978) The Physiological concepts in Indian medicine, Science and Philosophy of Indian medicine, Shree Beldyanath Ayurved Bhawan Ltd. [7] C. Chatfied (1996), Introduction to Multivariate Analysis, Chapman and Hall. [8] G. Coppin, A. Skrzyniarz (2003), Human centered processes : Individual and distributed decision support, IEEE Intelligent systems, pages 27 33. [9] D. Westwood, Flex reference guide, LPA, U.K [10] D. S. K. Mendis., A.S. Karunananda and U.Samarathunga (2004), Multi-Techniques Integrated tacit knowledge modelling system, International Journal of Information Technology, Vol 9, pp 265-271 [11] D.S.K Mendis, A.S. Karunananda and U.Samarathunga (2004), An Expert system for analysing Aurvedic human constituents, Shamisha Journal of Ayurveda, Vol 1, pp 145-150