D. Nauck, U. Nauck, and R. Kruse, "Generating classification rules with the neuro-fuzzy system NEFCLASS," presented at the Biennial Conf. North Amer. Fuzzy Information Processing Soc. (NAFIPS'96), Berkeley, CA.  Home/Search   Document Details and Download   Summary   Related Articles   Check

....models include systems, which combine different techniques into one single computational model. They share data structures and knowledge representations. Using ANN and FL, some of the major woks in this area are, Adaptive Network based Fuzzy Inference System (ANFIS) 4 5] NEFCON [6] NFCLASS [7], NEFPROX [42] FUN [8] SONF1N [9] F1NEST [10] FuNN [51] EfuNN [49] dmFuNN [50] and many others [53 54] 58] 63 64] Common to these approaches is their network like architecture, which is often used, in one way or the other, on a multi layered fuzzy rule base evaluation scheme ....

....possible rules need be considered, then only these rules are put into the network. A similar situation is applicable to Incremental leaming rule, where by if prior knowledge is available, then rule leaming does not need to start from scratch. 6. 3 NEFCLASS (Neuro Fuzzy Classification) NEFCLASS [7] [40 41] is used to derive fuzzy rules from a set of data that can be separated in different crisp classes. The rule base of a NEFCLASS system approximates an (unknown) function q0 that represents the classification problem and maps an input pattem x to its class Ci: 10 f x c C i go: R n ....

[Article contains additional citation context not shown here]

Nauk D, Nauck U and Kruse R, Generating Classification Rules with Neuro-Fuzzy System NEFCLASS, In proceedings of the Biennial Conference of the North American Fuzzy Information Processing Society (NAFIPS '96, Berkeley, 1996.

....by different methods give some idea about their relative merits. The number of rules and conditions does not characterize fully the complexity of the set of rules, since fuzzy rules have additional parameters. Else condition is not counted as a separate rule. The neurofuzzy NEFCLASS system [70] belongs to the best of its kind and if it had used context dependent linguistic variables it would probably achieve better results, but following the crowd the authors used three equally distributed fuzzy sets for each feature. The best 7 fuzzy rules classified correctly 96.7 IEEE TRANSACTIONS ....

....OF RULE EXTRACTION RESULTS FOR THE IRIS DATASET. F=FUZZY, C=CRISP,R=ROUGH,W=WEIGHTED. Method Rules cond. Type Reclassification features accuracy ReFuNN [10] 9 26 4 F 95.7 ReFuNN [10] 14 28 4 F 95.7 ReFuNN [10] 104 368 4 F 95.7 Grobian [72] 118 4 R 100.0 GA NN [65] 6 6 4 W 100.0 NEFCLASS[70] 7 28 4 F 96.7 NEFCLASS[70] 3 6 2 F 96.7 FuNe I[74] 7 3 F 96.0 C MLP2LN 2 2 1 C 95.7 C MLP2LN 2 2 2 C 96.0 C MLP2LN 2 3 2 C 98.0 SSV 2 2 2 C 98.0 of data. The system is not able to reduce the number of features automatically, but if used with the last two iris features it will give the ....

[Article contains additional citation context not shown here]

D. Nauck, U. Nauck and R. Kruse, "Generating Classification Rules with the Neuro--Fuzzy System NEFCLASS". Proc. Biennial Conf. of the North American Fuzzy Information Processing Society (NAFIPS'96), Berkeley, 1996

....adding new nodes if classification accuracy becomes too low. It may seem that neurofuzzy systems should have advantages in application to rule extraction, since crisp rules are just a special case of fuzzy rules. Quite many neurofuzzy systems are known and some indeed work rather well [42] 46] [47], 48] 49] However, there is a danger of overparametrization of such systems, leading to difficulty of finding optimal solutions [10] 50] even with the help of genetic algorithms or other global optimization methods. Systems based on rough sets [11] require additional discretization ....

Nauck D, Nauck U, Kruse R, "Generating Classification Rules with the Neuro-Fuzzy System NEFCLASS". Proc. Biennial Conf. of the North American Fuzzy Information Processing Society (NAFIPS'96), Berkeley, 1996.

....by learning, or by a combination of both. If a learning algorithm is applied that uses local information and causes local modifications in a fuzzy system, this approach is usually called neuro fuzzy system [7] We have already presented two neuro fuzzy approaches NEFCON [5] and NEFCLASS [6, 8]. The first one is used for control applications, and is trained by reinforcement learning based on a fuzzy error measure. The second one is used for classification of data, and is based on supervised learning. Both models can do structure and parameter learning by using a learning procedure ....

Detlef Nauck, Ulrike Nauck, and Rudolf Kruse. Generating classification rules with the neuro--fuzzy system NEFCLASS. In Proc. Biennial Conference of the North American Fuzzy Information Processing Society NAFIPS'96, pages 466--470, Berkeley, June 1996. IEEE.

....depending on the current error. The learning result might look like the situation in Fig. 4. x y Figure 4: Situation after training the classifier, i.e. modifying the membership functions To obtain an interpretable classifier some restrictions can be specified by the user. The NEFCLASS software [16] allows to impose the follwing restrictions on the learning algorithm: ffl a membership function must not pass on of its neighbors, ffl a membership function may be asymetrical, ffl membership functions must intersect at 0.5. How these restrictions influence the learning procedure is described ....

Detlef Nauck, Ulrike Nauck, and Rudolf Kruse. Generating classification rules with the neuro-fuzzy system NEFCLASS. In Proc. Biennial Conference of the North American Fuzzy Information Processing Society NAFIPS'96, Berkeley, CA, June 1996.

....for readability. Learning becomes more difficult but overfitting might not occur that easily and the learning result can be interpreted more easily. 4. A Simple Example To illustrate the effect of rule weights we use a simple example. We apply our neuro fuzzy learning environment NEFCLASS [11] [12] to the well known Iris data set [3] NEFCLASS is available for download from our homepage at http: fuzzy.cs.uni magdeburg.de. We chose the Iris example, because it is very simple and therefore suitable for demonstrating the effect of rule weights. The learning result shown here is of course not ....

Detlef Nauck, Ulrike Nauck, and Rudolf Kruse. Generating classification rules with the neuro-fuzzy system NEFCLASS. In Proc. Biennial Conference of the North American Fuzzy Information Processing Society NAFIPS'96, Berkeley, CA, June 1996.

No context found.

D. Nauck, U. Nauck and R. Kruse (1996b). Generating Classification Rules with the Neuro--Fuzzy System NEFCLASS. In Proc. Biennial Conference of the North American Fuzzy Information Processing Society NAFIPS'96, Berkeley.

....value (e.g. x 1 is positive big ) there is only one representation as a fuzzy set. It cannot happen that two fuzzy sets that are identical at the beginning of the learning process develop differently, and so the semantics of the rule base encoded in the structure of the network is not affected [7]. Connections that share a weight always come from the same input unit. 3 Learning in NEFCLASS A NEFCLASS system can be build from partial knowledge about the patterns, and can be then refined by learning, or it can be created from scratch by learning. A user has to define a number of initial ....

.... W (x 0 ; R) might be shared by other connections, and in this case might be changed more than once) iv) If an epoch was completed, and the end criterion is met, then stop; otherwise proceed with step (i) A discussion and an interpretation of the learning procedure of NEFCLASS can be found in [7] and in the accompanying paper at this conference in [6] 4 Pruning a Rule Base The learning algorithm of NEFCLASS provides good results for many classification problems. An example is given in the accompanying paper at this conference [6] However, a good interpretation of the learning result ....

[Article contains additional citation context not shown here]

Detlef Nauck, Ulrike Nauck, and Rudolf Kruse. Generating classification rules with the neurofuzzy system NEFCLASS. In Proc. Biennial Conference of the North American Fuzzy Information Processing Society NAFIPS'96, Berkeley, CA, June 1996.

No context found.

D. Nauck, U. Nauck, and R. Kruse, "Generating classification rules with the neuro-fuzzy system NEFCLASS," presented at the Biennial Conf. North Amer. Fuzzy Information Processing Soc. (NAFIPS'96), Berkeley, CA.

No context found.

D. Nauck, U. Nauck, and R. Kruse, "Generating classification rules with the neuro-fuzzy system NEFCLASS," presented at the Biennial Conf. North Amer. Fuzzy Information Processing Soc. (NAFIPS'96), Berkeley, CA.

No context found.

D. Nauck, U. Nauck, and R. Kruse, "Generating classification rules with the neuro-fuzzy system NEFCLASS," presented at the Biennial Conf. North Amer. Fuzzy Information Processing Soc. (NAFIPS'96), Berkeley, CA.

No context found.

D. Nauck, U. Nauck, and R. Kruse, "Generating classification rules with the neuro-fuzzy system NEFCLASS," presented at the Biennial Conf. North Amer. Fuzzy Information Processing Soc. (NAFIPS'96), Berkeley, CA.

Online articles have much greater impact   More about CiteSeer.IST   Add search form to your site   Submit documents   Feedback  

CiteSeer.IST - Copyright Penn State and NEC