Intelligent Data Analysis

More and more data is being accumulated each day,resulting in an ever increasing desire to sensibly interpret this data.There is a clear distinction between applying simple calculations (e.g. calculating the mean,standard deviation or trend detection) and actually forming an understanding of data and being able to use this understanding for making predictions or detecting outliers.

Examples of advanced data analysis:

Predicting employee sickness recovery time and determining the right course to take in the reintegration process.

• Using AI learning techniques, high dimensional data with unknown covariance can be used without preprocessing. Data is interpreted more accurately and better generalizing to new patients.
• Patient recovery time can be estimated more accurately than by using standard techniques.
• The influence of different approaches in the reintegration process can be estimated more accurately than a human expert could based on the given data.
• Occurence of work related sickness could be predicted at forehand or prevented by identifying main causes.

Stock management and pricing strategy based on rich information. A basic rule for any (internet based) shop is to attempt to have as little stock as is absolutely necessary and never to run out of stock completely. To achieve this, more than just the current sales records can be used to come to a more accurate prediction of future sales.

• Temporal sales patterns in similar products can be extracted.
• User feedback, internet reviews and website visitor information can be used to detect “hypes”.
• Respond to competitor pricing strategies by automatically monitoring websites.

Statistical Learning

Using a database of historical examples, statistical learning algorithms can automatically derive complex rules and make predictions on new data. One of the best known algorithms of this kind is neural network. Though the theory of neural networks dates back to the 1960s, correct and practical implementations of neural networks are surprisingly still rare. Furthermore, software developers remain skeptical to a technology they think of as a mysterious black box. In fact Neural Networks are far from mysterious; they rely on basic statistical principles and their performance can be easily measured and compared to basic analytic prediction methods. Neural Networks allow you to make predictions in (extremely) high-dimensional data patterns in milliseconds and with a higher accuracy than any traditional method.

Support Vector Machines (SVMs) are another example of a statistical learning algorithm which can provide better results depending on the type of the problem. Especially in text analysis, SVMs can prove to be very valuable.

Applications:

• Making predictions based on high-dimensional reference data
• Interpreting parameterized models
• Data compression

More on:
Artificial Neural Networks
Support Vector Machines

Evolutionary Algorithms

Finding an optimal parameter set for a given problem is often complicated by the inability to create an a-priori reference set containing “good” and “bad” examples. For example, this could be the case when the optimal solution isn’t yet known (not even for historical examples), or when a constant adjustment to new situations is required. In these cases, Neural Networks and SVMs provide insufficient flexibility and one can resort to algorithms inspired by evolutionary principles. In this case an optimal solution is found through genome (parameter) recombination, mutation and trial and error.

Applications:

• Solving high-dimensional problems
• Automatic creation of design variations

More on:
Evolutionary Algorithms