POSSIBILITIES
Dynamic simulation
Numerical solutions for (systems of) differential equations. From launching of a space missile to description of complex (bio ) chemical processes. May serve to make predictions, and supports development of models which help to understand basic mechanisms.
Regression techniques
Fitting data to linear, non-linear or dynamic models. Simple models, multi response models, multivariate models, time-related models. Estimation of most likely values for the parameters in a model, and their accuracies. Often least square optimization is used. For cases where measurement errors are significant in both predictor and response variables, orthogonal regression is the method of choice.
Shelf life estimation
An application of a multi-response non-linear regression model. In the example graph, loss of quality of a product can be described by a 1st order kinetic model (exponential decay equation). Stability is measured at different temperatures. The curves can be related by the Arrhenius equation. Integrated (simultaneous) analysis of all three curves allows prediction of shelf life for any temperature regime.
Monte-Carlo simulation
Generation of (millions of) random number is the basis of this method. Can be used for prediction of complicated processes, when mechanisms are more or less known, but subject to uncertainties. The weather forecast is a typical example of an application of Monte-Carlo simulation. But there are many more areas where stochastic prediction is more useful than deterministic prediction, for instance: business or financial forecasts, worst-case versus best case scenario analysis.
Trend Analysis
Most processes in real life show smaller or larger variations. It can be difficult to recognize gradual changes against the background of ‘natural’ variation. Ability to detect step changes as result of (unnoticed) changes in external conditions can be very valuable. In addition to step changes, also changes in trend can be important to recognize.
Cluster Analysis
Bringing structure in smaller or larger data sets can be very helpful for the recognition of patterns, fishing out relevant cases, identifying relationships. Clustering can be agglomerative (grouping together) of divisive (splitting up). The number of clusters can be pre-determined or can follow from the data. The separation between clusters can be sharp or be fuzzy.
Data handling, data mining, big data
Availability of massive amounts of data is more and more common these days. This calls for effective methods of handling these data; methods that take a step further than what is possible in a simple spread sheet program. A palette of techniques is available to get the crux out of large data sets. For instance, fishing out the differences between “good” and “bad” performance. Multivariate techniques like stepwise regression, principal component analysis, and partial least squares may reveal complex relationships.
Contact us to see whether we can shine a light on your data ...