Decision tree

At run time, a decision tree asks a series of questions in order, and based on the outcome of these delivers a label. In essence then, this is no more than series of nested if then else constructions in a computer program. For our cases, one set of questions might be if the token is lead and the following token is contains the string poison and if the token pollution or heavy is also foimd in the sentence then mark this as lead metal . 

Any series of questions can be formulated as a tree, and in fact decision trees can be written by hand. We do however have a number of algorithms which automatically build trees and it is 

Create an initial group containing all the instances of one ambiguous token. 

Design a set of questions based on our features which we think will help in the word identification 

Find the question which gives the biggest reduction in impurity 

It is relatively easy to apply the decision tree predicting function, so that even large trees are evaluated quickly. Starting at the root, the internal nodes are visited according to their decision rules If the decision rule at is fulfilled, then node V2fc+i is visited next, otherwise node V2fc+2- If a terminal node is reached, its function value is returned, otherwise the next decision rule is processed. 

In constructing a decision tree, the learning set is successively partitioned into two disjoint subsets. Partitioning is done according to a binary decision rule Xj a. Let be the index set of the observations represented by V. Further, 

Further details such as stopping criteria for growing decision trees, strategies for pruning decision trees and for constructing classification trees are described in [36]. An example of a classification tree to detect the presence of bromine in a mass spectrum is shown on pages 205-215. We will consider similar problems in Chapter 8, using an implementation by B. D. Ripley ([251], Chapter 7) via an interface to the statistics 

An interesting further development of regression trees is the software CUBIST [239], which combines recursive partitioning with linear regression. The predicting functions are RT with LM at terminal nodes. In [40] the aqueous solubility of compounds is modeled using this method. 

In general though, we have seen that it is intractable to assume that every combination of features is unique and needs to be calculated separately we simply never have enough training data to estimate such models accurately. One popular solution to tiiis is to use decision trees. These model interactions between features but do so in a particular way so as to concentrate on learning the influence of the most important features first, such that rare or undiscriminating feature combinations can be ignored with some safety. We 

---

Keywords reducing uncertainty, cost-effective information, ranking sources of uncertainty, re-processing seismic, interference tests, aquifer behaviour, % uncertainty, decision tree analysis, value of information, fiscal regime, suspended wells, phased development. 

There are two types of nodes in the decision tree decision nodes (rectangular) and chance nodes (circular). Decision nodes branch into a set of possible actions, while chance nodes branch into all possible results or situations. 

The decision tree can be considered as a road map which indicates the chronological order in which a series of actions will be performed, and shows several possible courses, only one of which will actually be followed. 

In this example it would therefore be justifiable to spend up to 22 million on appraisal activity which would distinguish between the high, medium, and low STOMP cases. If it would cost more than 22 million to determine this, then it would be better to go ahead without the appraisal. The decision tree has therefore been used to place a value on the appraisal activity, and to indicate when it is no longer worthwhile to appraise. 

The benefit of using the decision tree approach is that it clarifies the decision-making process. The discipline required to construct a logical decision tree may also serve to explain the key decisions and to highlight uncertainties. 

The following decision tree shows a logical sequence of decisions (shown in the rectangular boxes) and chance outcomes (chance events are represented by circles). At each decision point, petroleum economics is applied to determine the choice, with the criterion being to achieve a positive EMV. 

To become familiar with the structure and task of decision trees... 

Kohonen network Conceptual clustering Principal Component Analysis (PCA) Decision trees Partial Least Squares (PLS) Multiple Linear Regression (MLR) Counter-propagation networks Back-propagation networks Genetic algorithms (GA)... 

Figure 9-2. Decision tree for classifying stellar spectra. ...

Classification describes the process of assigning an instance or property to one of several given classes. The classes are defined beforehand and this class assignment is used in the learning process, which is therefore supervised. Statistical methods and decision trees (cf. Section 9.3) are also widely used for classification tasks. 

Decision trees give a graphical representation of a procedure for classification. They consist of nodes and branches the leaf nodes give the classification of an instance. 

A series of monographs and correlation tables exist for the interpretation of vibrational spectra [52-55]. However, the relationship of frequency characteristics and structural features is rather complicated and the number of known correlations between IR spectra and structures is very large. In many cases, it is almost impossible to analyze a molecular structure without the aid of computational techniques. Existing approaches are mainly based on the interpretation of vibrational spectra by mathematical models, rule sets, and decision trees or fuzzy logic approaches. 

The two principal approaches of interest for situations involving risk and uncertainty are decision trees and Monte Carlo simulation (26,29,30). 

Decision Trees In a typical decision tree, illustrated in a very simplified form by Fig. 9-24, each node represents a decision point (DP) at which one or more alternatives are available. Some quantifiable result of each alternative is chosen as a basis for comparison for example, the net present value (NPV). A value is assigned to the probability of attaining each result, either cumulative or not as required. These may be obtained by the methods just described or otherwise. The estimates are subject to the restriction that the sum of the proba-... 

FIG. 9-24 Effect of decision-tree options on net present value. 

Figure 5 is an example of a decision tree you may find useful when considering QRA for particular process safety applications. The decision tree illustrates a flowchart of questions you can ask yourself (or others) to decide how far through the process of risk analysis to go to satisfy a need for increased risk understanding. 

Decision trees are not used in this book since they are most useful when targeted to a specific process attempts to generate comprehensive matrices rapidly lead to extremely complex schemes. Instead, the book should be used to help generate suitable matrices or to supplement the decision-making steps in published matrices such as [ 199 ]. Many of the decision steps, such as the conditions under which discharges of some specified effective energy may occur, are not properly understood and continue to be controversial. 

A significant development of the study was the use of event trees to link the system fault trees to (lie accident initiators and the core damage states as described in Chapter 3. This was a response to the ditficulties encountered in performing the in-plant analysis by fault trees alone. Nathan Villalva and Winston Little proposed the application of decision trees, which was recognized by Saul Levine a.s providing the structure needed to link accident sequences to equipment failure. 

CBDTM Cause Based Decision Tree Method Singh et al, 1993... 

Event trees are adaptations of decision trees used for analyzing the risk of financial decisions. 

A decision tree for Design Methodology is illustrated in Fig. 3.2. Each step in the tree is explained briefly below. The steps have also their own subtrees, which are described separately. 

This task consists of defining target levels for indoor and outdoor conditions based on requirements for laws and regulations, human health, production processes and equipment, and type of premises and construction. Target levels should also be defined for the ventilation system. For the decision tree, see Fig. 3.6. 

Based on technical specifications, acceptable equipment is identified. The final selection is made on the same basis as in Selection of System. The decision tree is shown in Fig. 3.13. 

Brodley, C.E. and Utgoff, P.E., 1995. Multivariate Decision Trees. Machine Learning, 19, 45. 

Fayyad, U.M. and Irani, K.B., 1992. On the handling of Continuous-Valued Attributes in Decision Tree Generation. Machine Learning, 8, 87. 

Mingers, J., 1989. An Empirical Comparison of Pruning Methods for Decision Tree Induction. Machine Learning, 4, 221. 

Quinlan, J.R., 1990. Decision Trees and Decision making. IEEE Trans Systems Man Cybernetics, 20(2), 339. 

The calculation method can be selected by application of the decision tree in Figure 9.2. The liquid temperature is believed to be about 339 K, which is the temperature equivalent to the relief valve set pressure. The superheat limit temperatures of propane and butane, the constituents of LPG, can be found in Table 6.1. For propane, T, = 326 K, and for butane, T i = 377 K. The figure specifies that, if the liquid is above its critical superheat limit temperature, the explosively flashing liquid method must be chosen. However, because the temperature of the LPG is below the superheat limit temperature (T i) for butane and above it for propane, it is uncertain whether the liquid will flash. Therefore, the calculation will first be performed with the inclusion of vapor energy only, then with the combined energy of vapor and liquid. 

---