Scaling of Data

From an industrial perspective, quantitative knowledge of the existence of different transporters within the cellular system used in screening procedures is of major importance as it can influence both the predictive value of the permeability coefficients and interpretation of the results. In addition, information on species differences or similarities or discrepancies between cell culture models and animals now provide an important basis for the scaling of data during the early phases of drug discovery for animals or humans [48]. 

In data analysis, data are seldom used without some preprocessing. Such preprocessing is typically concerned with the scale of data. In this regard two main scaling procedures are widely used zero-centered and autoscaling. 

SOLVE/RESOLVE is a program system that permits automation of all the steps between processed data and interpretation of phased maps. These include scaling of data measured at multiple wavelengths, location of anomalous scatterers. 

Example 20 Scaling of Data An experimental batch drying curve has been measured at 100°C, and drying time was 2 h. Estimate the drying time at la) 100°C and (b) 150°C for (i) a fluidized-bed dryer and (ii) a conical vacuum dryer at 100-mbar absolute pressure, for a batch size 10 times greater than that of the test. Assume for the fluidized bed that temperature driving forces are proportional to and batch drying time is proportional to bed depth, and for... 

Inadequate temporal scale of data routine monitoring is at regular time intervals, whilst bulk sediment transport is highly skewed to high flows. 

A disadvantage of measures based on the Lp metric is their dependence on the dimensions used. Scaling of data is frequently unavoidable if these distance measures are to be applied. 

Scaling of data is not necessary if the Mahalanobis distance is used. In addition, with this measure distortion occasioned by correlations of features or feature groups is avoided. In contrast, if the Euclidean distance were applied in the case of two highly correlated variables, these variables would be used as two independent features although they provide identical information. 

When contemplating the scaling of data, one has to consider what is the limitation to the performance on the test machine, because different limitations require different scaling techniques. Moreover if the scale-up factor is large, then the scaling may introduce another limitation, if the scale factor for the second limitation is smaller than the first. 

Through analysis of the effect of fitting various time scales of data for predicting long term behavior, it was found that for the 700/100°C gradient tests, the predictions of the 7,500 and 12,500 hour fits were the best. For the lower temperature gradient tests, results varied. 

Another variant of spherical PCA that accounts for various scales of data parameters is the so-called elliptical principal component analysis (ePCA). In contrast to spherical PCA, in the ePCA algorithm the data samples are projected onto a hyperellipse. The radii of the hyperellipse are the scales of parameters expressed by the robust Qn scale estimator. 

---