Errors in data processing

Numerical differentiation should be avoided whenever possible, particularly when data are empirical and subject to appreciable observation errors. Errors in data can affect numeric derivatives quite strongly i.e., differentiation is a roughening process. When such a calculation must be made, it is usually desirable first to smooth the data to a certain extent. 

The type of data collected on human error and the ways in which these data are used for accident prevention will vary depending upon the model of error and accident causation held by the management of an organization. This model will also influence the culture in the plant and the willingness of personnel to participate in data collection activities. In Chapters 1 and 2 a number of alternative viewpoints or models of human error were described. These models will now be briefly reviewed and their implications for the treatment of human error in the process industry will be discussed. 

Ti measurements were performed at 250, 275 and 300 K by inversion-recovery (7r-T-7r/2-5T i) sequences on a JE0L-FX-100 and a Bruker WP-80 spectrometers. On this latter the "repetitive frequency shift" method of Brevard et al. (18) was used, where two systematic instrumental errors (drift, round off errors in FT processing.. .) are uniformly distributed through all data points. The NOE measurements are reproducible within 10-20 %, while the average standard error on the Tj values is of about 5 %. 

Finding the source location and the time history of the solute in ground-water can be categorized as a problem of time inversion. This means that we have to solve the governing equations backward in time. Modeling contaminant transport using reverse time is an ill-posed problem since the process, being dispersive is irreversible. Because of this ill-posedness, the problems have discontinuous dependence on data and are sensitive to the errors in data. 

Error Reduction Without automation, data needs to be entered and reentered at several stages in the process. This inevitably introduces clerical error. With electronic data exchange, the data is only entered once and reused in all subsequent steps. Thus, human error in data entry can be significantly reduced. 

These claims were immediately challenged through molecular dynamics simulation by Chialvo and Cummings [65] and by Loffler et al [131], and at once it became clear that the NDIS-93 data were seriously flawed. Given the complexity of the NDIS experiments and the numerous possible sources of errors in the processing of the raw data, it was reasonable to suspect the presence of artifacts [131]. Yet, initially it was also unclear whether the disagreement between the predicted and measured correlation functions was a reflection of unrealistic intermolecular models, inadequate methods of processing the diffraction raw data, or a combination of both factors. 

The use of SVM cannot solve all problems of noise in data processing, but it is possible to use SVM technique to improve noisy data processing in many ways. For example, it can provide some ways for outlier deleting By leave-one-out (LOO) cross-validation method, we can delete the data samples with large error in prediction, and make the improvement of data files. Besides, the adoption of e-insensitive loss function in support vector regression makes it more robust to noisy data sets. 

Pickering, very rightly, points to several questions concerning temporality, representation, and the ageney of material matter. He also proposes the mangle metaphor as the means to describe and understand the process of science wherein both human and non-human actors take part. It is a process where scientists make apparatus, which in turn makes data, in a continuous process of dialectical trial and, most often, error. In this process both the human (the scientist) and the non-human actant (the apparatus) are active and consequently have agency. 

While many methods for parameter estimation have been proposed, experience has shown some to be more effective than others. Since most phenomenological models are nonlinear in their adjustable parameters, the best estimates of these parameters can be obtained from a formalized method which properly treats the statistical behavior of the errors associated with all experimental observations. For reliable process-design calculations, we require not only estimates of the parameters but also a measure of the errors in the parameters and an indication of the accuracy of the data. 

In many process-design calculations it is not necessary to fit the data to within the experimental uncertainty. Here, economics dictates that a minimum number of adjustable parameters be fitted to scarce data with the best accuracy possible. This compromise between "goodness of fit" and number of parameters requires some method of discriminating between models. One way is to compare the uncertainties in the calculated parameters. An alternative method consists of examination of the residuals for trends and excessive errors when plotted versus other system variables (Draper and Smith, 1966). A more useful quantity for comparison is obtained from the sum of the weighted squared residuals given by Equation (1). 

When designing and evaluating an analytical method, we usually make three separate considerations of experimental error. First, before beginning an analysis, errors associated with each measurement are evaluated to ensure that their cumulative effect will not limit the utility of the analysis. Errors known or believed to affect the result can then be minimized. Second, during the analysis the measurement process is monitored, ensuring that it remains under control. Finally, at the end of the analysis the quality of the measurements and the result are evaluated and compared with the original design criteria. This chapter is an introduction to the sources and evaluation of errors in analytical measurements, the effect of measurement error on the result of an analysis, and the statistical analysis of data. 

Rosenberg, J., R.S.H. Mah, and C. lordache, Evaluation of Schemes for Detecting and Identifying Gross Errors in Process Data, Indushial and Engineeiing Chemistiy, Reseaieh, 26(.3), 1987, 555-564. (Simulation studies of various detection methods)... 

The composite envelope is then plotted over the envelope of each individual peak. It is seen that the actual retention difference, if taken from the maxima of the envelope, will give a value of less than 80% of the true retention difference. Furthermore as the peaks become closer this error increases rapidly. Unfortunately, this type of error is not normally taken into account by most data processing software. It follows that, if such data was used for solute identification, or column design, the results can be grossly in error. 

Three major themes have been emphasized in this chapter. The first is that an effective data collection system is one of the most powerful tools available to minimize human error. Second, data collection systems must adequately address underlying causes. Merely tabulating accidents in terms of their surface similarities, or using inadequate causal descriptions such as "process worker failed to follow procedures" is not sufficient to develop effective remedial strategies. Finally, a successful data collection and incident investigation system requires an enlightened, systems oriented view of human error to be held by management, and participation and commitment from the workforce. 

---