Improving quantitative model performance

Once one builds a quantitative model and assesses its performance using either the validation methods discussed above or actual on-line implementation, the unavoidable question is Can I do better In many cases, the answer is quite possibly. There are several different actions that one could take to attempt to improve model performance. The following is a list of such actions, which is an expansion of a previously published guide by Martens and Naes.1 

This means for improvement concerns the experimental procedures that are used to collect and analyze the calibration samples. In PAC, sample collection can involve either a highly automated sampling system, or a manual sampling process that requires manual sample extraction, preparation, and introduction. Even for an automated data collection system, errors due to fast process dynamics, analyzer sampling system dynamics, non-representative sample extraction, or sample instability can contribute large errors to the calibration data. For manual data collection, there are even more error sources to be considered, such as non-reproducibility of sample preparation and sample introduction to the analyzer. 

With regard to analysis, there could be a different set of operating parameters for the analyzer instrument that would result in improved calibration data. For example, for an FTIR spectrometer, one could decrease the spectral resolution to obtain a better spectral S/N for the same total scan time. Such a strategy could be advantageous if the spectral features being used for calibration have relatively low resolution. There could also be specific sets of instrument-operating parameters where the instrument response is noisy or unstable over time, in which case they can be changed to improve precision and stability of the analyzer data. 

There are several strategies and tools that can be used to assess, and subsequently minimize, such sampling-based errors in the calibration data. The use of replicate samples in the calibration experiment can help to detect noise issues around instrument 

This improvement strategy involves the reduction of irrelevant information in the X-data, thus reducing the burden on the modeling method to define the correlation with the Y-data. Various types of X-data pre-processing, discussed earlier (Section 8.2.5), can be used to reduce such irrelevant information. Improvement can also be obtained through elimination of X-variables that are determined to be irrelevant. Specific techniques for the selection of relevant X-variables are discussed in a later section. 

---

To develop and apply assumption-free learning frameworks and methodologies, aimed at uncovering and expressing in adequate solution formats performance improvement. . opportunities, extracted from existing data which were acquired from plants that cannot be described effectively through first-principles quantitative models. 

In another work, Parra and coworkers proposed a method based on chemically modified voltammetric electrodes for the identification of adulterations made in wine samples, by addition of a number of forbidden adulterants frequently used in the wine industry to improve the organoleptic characteristics of wines, like, for example, tartaric acid, tannic acid, sucrose, and acetaldehyde (Parra et ah, 2006b). The patterns identified via PCA allowed an efficient detection of the wine samples that had been artificially modified. In the same study, PLS regression was applied for a quantitative prediction of the substances added. Model performances were evaluated by means of a cross-validation procedure. 

At this point it should be stressed that the application of quantitative modeling using the TP Model and EXAFS spectroscopy as well as the performance of calculations based on DFT may lead to an improvement or to a slight modification of the tentative local structures proposed in this work. 

The numerical accuracy of simulations performed using this model is affected by several factors. These include a) the degree of triangulation, b) the number of marching steps taken along the flow direction and c) the order of the polynomial basis function. Numerical accuracy improves as a, b and c increase, however the computational time can become excessive. Therefore, it was necessary to quantitatively determine the effects of these variables on numerical accuracy. 

It is clear that quantitative predictions of research performance, and models of possible improvements, are not well accepted by research managers. Two major barriers are the uncertainty that exists at early stages of R D and the mix of repetition and variation between different projects. 

Process modeling is usually performed for two reasons. For fundamental and scientific studies a process model serves to explain and predict the quantitative behavior of physical or chemical phenomena in the process. The predictive capability of the model, however, is usually exploited by the engineer in order to improve the process. Once the model, the process and problem limitations, and the criterion for improvement are clearly and... 

---