Rule extraction

As a rule, extraction is much more efficient from acid than from neutral salt or alkaline aqueous solutions (Fig. 4). 

Rule extraction from a mutagenicity data set using adaptively grown phylogenetic-like trees../. Chem. Inf. Comput. Sci. 42, 1104—1 111. 

Current trends in neural networks favor smaller networks with minimal architecture. Two major advantages of smaller networks previously discussed are better generalization capability (8.4) and easier rule extraction (13.2). Another advantage is better predictive accuracy, seen when a large network is replaced by many smaller networks, each for a subtask or a subset of data. A typical example is the protein classification problem, where n individual networks can be used to classify n different protein families and increase the prediction accuracy obtained by one large network with n output units. The improvement is especially significant when there is sufficient data for fine-tuning individual neural networks to the particularity of the data subsets. The use of ensembles of small, customized neural networks to improve predictive accuracy has been shown in numerous cases. 

Fig. 10 Rule extraction from neurofuzzy model. Values in brackets are confidence levels.

Johansson, U., Konig, R. and Niklasson, L. (2003) Rule extraction from trained neural networks using genetic programming. 13th International Conference on Artificial Neural Networks, Istanbul, Turkey, supplementary proceedings, pp. 13-16. 

An Example Using Genetic Programming-Based Rule Extraction... 

The example described here employs genetic programming (GP) (see further description of the method in Section 14.3.2.2) and the genetic rule extraction (G-REX) algorithm [54, 55]. 

Bacha, P.A., Gruver, H.S., Den Hartog, B.K., Tamura, S.Y. and Nutt, R.E. (2002) Rule extraction from a mutagenicity data set using adaptively grown phylogenetic-like trees. J. Chem. Inf. Comput. Sci., 42, 1104-1111. 

As a general rule, extraction is best when the density of the compressed gas approaches that of the liquid. Often there is a large increase of solubility as the pressure increases at pressures just above the critical pressure. Further increased pressure then results in little increased solubility. 

Figure 6 shows a structure of applying fuzzy logic in control. First, two types of inputs must be obtained numerical inputs and human knowledge or rule extraction from data (i.e., fuzzy rules). Then the numerical inputs must be fuzzified into fuzzy numbers. The fuzzy rules consist of the fuzzy membership functions (knowledge model) or so-called fuzzy associative memories (FAMs). Then the... 

Human ICnowledge or Rule Extraction from Data... 

QianY, Liang J, Dang C. Converse approximation and rule extraction from decision tables in rough set theory. Comput Mathematics 2008 55 1754—1765. 

Another important factor that influences on the accuracy of models is a metabolic pathway or metabohc activation of the parent compound which is not usually considered. It would be reasonable to predict the first reactive metabolite or find an actnal proof that it is produced, and then make a separate prediction for the metabolite of interest. Several computer programs provide either the prediction of htunan metabolism or common metabolic pathway based on the rules extracted from several species of mammals (rat, mouse, hitman, hamster), using a mixture of in vitro and in vivo metabohsm data of compoimds [68, 89]. However, in these programs, predictions made for the possible reactive metabolites do not affect the prediction for a parent compoimd. 

In this rule (as well as in all important kinds of logical rules extracted from data), three parts with different semantics can be distinguished ... 

Most frequently, the formal rules extracted from artificial neural networks have the form ... 

Over the last two decades, various rule-extraction methods have been proposed for trained neural networks, but so far none of them has... 

Examples of two-dimensional cuts visualising the unions of polyhedra that determine the antecedents of two combined-form rules extracted from the same trained MLP are shown in Figure 6.8 and Figure 6.9. Both cuts correspond to identical pairs of input variables, but to the increasingly restrictive ou ut conditions. 

Figure 6.9. A two-dimensional cut of the union of polyhedra from the antecedent of a combined-form rule extracted from the same MLP as in Figure 6.8. This cut corresponds to the same input variables as the cut in Figure 6.8, but for a more restrictive consequence propene yield > 9% .

In the rule-extraction method outlined above, the possibility of replacing a polyhedron P with a rectangular area Ri is assessed according to the following principles ... 

Conjunctive-form rules are also very convenient from the visualisation point of view — since cuts of rectangular areas eoincide with the corresponding projections of those areas, the values of no variables need to be fixed. As an example. Figure 6.10 shows three-dimensional cuts determining the antecedents of conjunctive-form rules extracted from a trained MLP with six input neurons and two output neurons, assuming the above interpretation of the variables to which those neurons correspond. The rules are extracted according to the principles (i)-(iii) for the consequence propene yield > 8% and are listed in Table 6.2. 

Table 6.2. Antecedents of the conjunctive-form rules extracted from a trained MLP using the method described in this section for the consequence propene yield > 8% .

Holena, M. (2006). Piecewise-linear neural networks and their relationship to rule extraction from data. Neural Comput, 18, 2813-2853. 

---