Clustering of data

This method tends to create a cluster of data near the origin as shown in Figure E.2.2. 

Cluster analysis is similar in concept to pattern recognition. It can define the similarity or dissimilarity of observations or can reveal the number of groups formed by a collection of data. The distance between clusters of data points is defined either by the distance between the two closest members of two different clusters or by the distances between the centers of clusters. 

Clustering of data items (unsupervised learning) in classes, after establishing class limits inductively from existing data sets... 

The third cluster of data points defines a particularly interesting d-SoC, lucid dreaming. This is the special kind of dream named by the Dutch physician Frederick van Eeden 88 or 115, ch. 8, in which you feel as if you have awakened in terms of mental functioning within the dream world you feel as rational and in control of your mental state... 

Analysis of variance in general serves as a statistical test of the influence of random or systematic factors on measured data (test for random or fixed effects). One wants to test if the feature mean values of two or more classes are different. Classes of objects or clusters of data may be given a priori (supervised learning) or found in the course of a learning process (unsupervised learning see Section 5.3, cluster analysis). In the first case variance analysis is used for class pattern confirmation. 

Perceptions, multilayer perceptions and radial basis function networks require supervised training with data for which the answers are known. Some applications require the automatic clustering of data, data for which the clusters and clustering criteria are not known. One of the best known architectures for such problems is the Kohonen selforganizing map (SOM), named after its inventor, Teuvo Kohonen (Kohonen, 1997). In this section the rationale behind such networks is described. 

Djmberg C, Svedhndh P, Nordblad P, Hansen MF, Bodker F, Morup S (1997) Dynamics of an interacting particle system Evidence of critical slowing down. Phys Rev Lett 79 5154-5157 Domany E (1999) Snperparamagnetic clustering of data— The definitive solution of an ill-posed problem. Physica A 263 158-169... 

Identification of novel biomarkers of toxicity Previously, the detection of novel biomarkers of toxic effect has mainly been serendipitous. However, it is now possible to use a combined NMR-expert systems approach to systematically explore the relationships between biofluid composition and toxicity and to generate novel combination biomarkers of toxicity. Pattern recognition maps can be examined for evidence of clustering of data according to site and type of toxic lesion. 

A scatterplot of the results, with outliers, removed, showed three groups of pottery based on their elemental composition (Fig. 8.13). The values used on the X- and T-axes of the scatterplot are based on principle components, a technique to create summary statistics that combine the results from all of the elements used in the study. Each data point on the graph represents a sherd sample from Pinson Mounds. Thus, the X- and T-axes use most of the results of the NAA measurements. The authors of the study then drew ovals around clusters of data points in the graph to distinguish three compositional groups. These ovals should encompass 90% of the data points in the group. These compositional groups should represent pottery... 

They represent an isolated cluster of data points on a 6 0 vs. 6 C plot and reveal slightly heavier carbon and lighter oxygen isotopic compositions than the other marine calcites. 

We can assign names of known d-SoCs to the three clusters of data points in the graph. Ordinary consciousness (for our culture) is shown in... 

Many tests exist for detecting outliers in univariate data, but most are designed to check for the presence of a single rogue value. Univariate tests for outliers are not designed for multivariate outliers. Consider Figure 1.6, the majority of data exists in the highlighted pattern space with the exception of the two points denoted A and B. Neither of these points may be considered a univariate outlier in terms of variable x or x2, but both are well away from the main cluster of data. It is the combination of the two variables that identifies the presence of these outliers. Outlier detection and treatment is of major concern to analysts, particularly with multivariate data where the presence of outliers may not be immediately obvious from visual inspection of tabulated data. 

Standard deviation (S.D.) is the square root of the variance that is a measure of the scatter of values about the mean value the smaller the value is, the more indicative of the tighter clustering of data about the mean. By calculating the mean and S.D. for a reference population, 1 S.D. will contain 68% of all values, 2 S.D. will contain 95.5% of all values, and 3 S.D. will contain 99.7% of all values. 

Fig. 11. oMEDA vector of two clusters of data from the 10 PCs PCA model of a simulated data set of dimension 100 x 100. This model captures 30% of the variability. Data present two clusters in variable 10. 

In Figure 11 an example of oMEDA is shown. For this, a 100 x 100 data set with two clusters of data was simulated. The distribution of the observations was designed so that both clusters had significantly different values only in variable 10 and then data was auto-scaled. The oMEDA vector clearly highlights variable 10 as the main difference between both clusters. 

These methods are aimed at projecting the original data set from a high-dimensional space onto a line, a plane, or a three-dimensional coordinate system. Perhaps the best way would be to have a mathematical procedure that allows you to sit before the computer screen pursuing the rotation of the data into all possible directions and stopping this process when the best projection, that is, optimal clustering of data groups, has been found. In fact, such methods of projection pursuits already exist in statistics and are tested within the field of chemometrics. 

If we were to locate a line by eye through a data set y, we would probably try to balance the distance of the line from the data points. We would take distances on the upper side of the line and balance them with distances on the lower side. If we were really good, we would take all data points into consideration, and maybe even decide that a longer distance from the line to that data point way up there is balanced by the many shorter distances to the cluster of data points on the lower side of the line. 

As a consequence of this difference, the samples analyzed by Eastin (1970) should form two clusters of data points on the Rb-Sr isochron diagram (not shown). The line connecting these data clusters is a mixing line whose slope cannot be used to date these rocks. 

---