Influence of the Training Data

The SVM possesses some desirable characteristics such as good generalization ability of the classifier function, robustness of the solrrtion, sparseness of the classifier and an automatic control of the solution complexity. Moreover, the formalism provides an explicit knowledge of the data poirrts (termed support vectors ), which are important in defining the classifier function. This feature allows an interpretation of the S VM-based classifier model in terms of the training data. Robustness of SVM is achieved by considering absolute, irrstead of qrradratic, values of the errors. As a consequence, the influence of outliers is less pronounced [26]. 

With all QSPR studies it is not possible to separate the influence of the data used to train the model and the computational approach used to derive the model from the final model. Ideally, the QSPR should be sufficiently general to be applied to any compound that is reasonably represented by the data used to derive the model. 

The selection of training data presented to the neural network influences whether or not the network learns a particular task. Some major considerations include the generalization/memorization issue, the partitioning of the training and prediction sets, the quality of data, the ratio of positive and negative examples, and the order of example presentation. 

Predictive models are built with ANN s in much the same way as they are with MLR and PLS methods descriptors and experimental data are used to fit (or train in machine-learning nomenclature) the parameters of the functions until the performance error is minimized. Neural networks differ from the previous two methods in that (1) the sigmoidal shapes of the neurons output equations better allow them to model non-linear systems and (2) they are subsymbolic , which is to say that the information in the descriptors is effectively scrambled once the internal weights and thresholds of the neurons are trained, making it difficult to examine the final equations to interpret the influences of the descriptors on the property of interest. 

To apply fractional factorial designs, the control variables have to be of nominal or ordinal scale or the relations to the performance measures are assumed to be linear. Here, the influence of the number of available trains, the number of working days for RTC handling, and the number of transfer arms is investigated. As status quo, the data provided in example 9 is used Prom Monday to Saturday trains can be dispatched and RTCs can be processed. For each product an equal number of transfer arms is available at both sites according to Table A. 12 in the appendix. As an alternative configuration, the... 

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... 

When forming a model, we wish to minimize the influence of random variations in the data and maximize the influence of the underlying causal effects. It is therefore important to use many (X, Y) pairs for training, as the standard error of a sample set is inversely proportional to the number of samples. Conversely, for estimation of the MSEP we need many (X, Y) test pairs, for exactly the same reasons. When forming a model, we usually have a limited number of (X, Y) pairs to use. There is therefore a trade-off between using samples for model formation and MSEP estimation. A number of methods have been suggested for addressing this problem and these will now be discussed. 

Theory-laden NOS. When asked to account for differences in scientists interpretations of the same data, 36% of the drama and 6% of the control group participants expressed informed conceptions of this NOS aspect. They acknowledged that scientists theoretical backgrounds and training influence their... 

---