Trufa – The Science Inside
All about AI, Advanced Statistics & more
Trufa is a SaaS analytics application using Artificial Intelligence (AI) to deliver opportunities for economic improvements in the enterprise. It combines multiple advanced statistical methods and state-of-the-art machine learning algorithms to discover and evaluate process optimization potentials. In the following whitepapers, Prof. Dr. Andreas Mielke describes some of these methods implemented in the Trufa application.
- #1 Robust Statistics
- #2 Maximum Entropy
- #3 Time Series Analysis, Modelling, and Forecast
- #4 Regression Outliers
- #5 Neural Networks
Whitepaper No 1: Robust Statistics
Statistics is known to everybody, simply because in our current live we often see statistical analyses. But what is robust statistics, why should statistics be robust? And why should we use robust statistics instead of well known usual statistics, when we analyze business data? The present paper answers this questions and explains why only robust statistics guaranties reliable results when we analyze business data. The aim is to provide some very basic understanding of the why and how of robust statistics. For details of the mathematical foundation or the implementation of robust procedures we refer to the literature mentioned at the end.
Whitepaper No 2: Maximum Entropy
Entropy is a quantity which appears in statistical physics. How is it connected to business data and why should it be maximized? The present paper tries to answer that question. We show that the maximum entropy principle ensures that the outcome of a statistical model is unbiased, robust, and actionable. The maximum entropy method is a general-purpose technique in machine learning. It has a simple and precise mathematical foundation. A number of aspects make it well suited for the modelling of distributions of business data.
Whitepaper No 3: Time Series Analysis, Modelling, and Forecast:
The paper aims at explaining how time series analysis can be used for the prediction of business data. We present which methods are available, which methods are suitable for business data, and how prediction of business data can be implemented. As an example, we will use liquidity prediction.
Whitepaper No 4: Regression Outliers
Standard regression often yields bad results if outliers are present. There are two ways out of this dilemma. The first is to remove the outliers. The second is to use robust methods for which the result does not depend on outliers. If the data set is large, the second possibility is often the only possible one. In this paper, we point out that robust methods can and should be used for outlier detection. The reason is that outliers often contain additional information and are thus important. They may for instance show that an additional factor is relevant to understand the behavior of the full data set. A precise outlier detection is therefore mandatory.
Whitepaper No 5: Neural Networks
Neural networks are a useful tool for simulations, classifications, predictions, and forecasts in various areas. In Trufa we use neural networks to determine the functional relationship between all kinds of quantities. The aim of this paper is to explain some of the main aspects of neural networks to the non-expert and to make clear where and how they are used in Trufa.
Whitepaper No 6: Explainable AI (XAI)
Users view Artificial Intelligence (AI) systems often as a black box. The system yields a result but no explanation of it or of the way it was obtained. In many applications it is important that the user understands the result of AI, because the explanation provides better trust and acceptance. In this paper, we explain what Explainable AI (XAI) is and how it is realized in Trufa.