Predictive Analytics is the third part of our Analytics series. This form of analysis requires as a starting point the model parameters object, dimension, value, and event, as described in the article What is Descriptive Analytics. Based on these parameters, scenarios can be identified that should be avoided or promoted. This identification process is the diagnostic analysis. Subsequently, statistical models can be created which, based on if-then relationships, make a prediction for subsequent events. This predictive analysis is not only suitable for future events, but also for identifying events from the past that are not yet known: For example, when an unknown culprit is identified on the basis of available data. It is also used in the energy industry in order to timely provide the necessary capacities for the predicted consumption.

The core of predictive analytics is to capture relationships between relevant variables of the current state and predicted variables from past events and to use them to predict an unknown outcome.

An example outside of intralogistics is the assignment of texts of unknown origin to known authors. For this purpose, the available data is processed to bring identifiable author-style properties and other data sources into a common context with clearly defined relationships between the variables, something the lifetime of an author and linguistic constructs known at that time. For example, it is unlikely that a 17th-century author would have used the phrase „It’s all Greek to me“. With predictive analysis and a data set prepared via descriptive analysis, texts whose author is not known can thus be assigned to a known author with a certain degree of probability. The more correlations that are made in the available data set, the higher the quality of this assignment.

Predictive analysis in intralogistics

A question we are often asked by distribution center employees is: „Why did things go well yesterday, but not today, even though the order situation is the same?

Predictive analytics can be used to find possible answers with assigned probabilities. The basis for this is a previously performed diagnostic analysis that precisely describes the possible states of the various systems and assigns them a qualitative value. In the previous article ‚Diagnostic Analytics‘ we used the example of an analysis tree for the delayed departure of a truck, which we will briefly refer to here.

Je nach Ergebnis der Daten, ergeben sich mögliche Erklärungsmodelle. Wie hier am Beispiel der langen Beladungsdauer.

After the diagnostic part, the predictive analysis starts and assigns a probability of occurrence to elements of this decision tree: Thus, with each applicable parameter, such as „high sickness rate“ or „large order“, the probability that the condition „truck loading takes too long“ occurs increases. What is applied to past data in the diagnostic analysis, predictive analysis projects to a still unknown or future state. This creates the necessary reaction time to still be able to react to a situation. The leading personnel is considerably supported in the prediction, since the individual experience with the business processes as a factor loses weight.

Complex event chains and short time frames as a risk for intralogistic processes

Two exemplary scenarios, which are to be recognized in time by predictive analysis, are the snowball and the butterfly effect. In the former, errors that initially appear uncritical develop into serious ones that, in the worst case, interrupt the entire process flow. With the latter, the connection between the same or similar initial situation and a significantly different further development of the system cannot be explained: although the order situation and the available resources are the same on two different days in a warehouse, on day A the system runs smoothly and on day B, contrary to expectations, malfunctions occur.

Can the flapping of a butterfly’s wings trigger a tornado in Texas?

– Edward N. Lorenz

Transferred to intralogistics, the situation in distribution centers is that disruptions accumulate even though the initial situation is not exceptional and has already been smoothly handled at another point in time. The connection between effect and causer is – yet – not known.

Predictive analysis can stop rolling snowballs before they develop into an avalanche, since the problematic system states and their relationships are recorded. Using the example from diagnostic analysis, a predictive path could look like this:

Wo die diagnostische Analyse die Zusammenhänge herstellt, parametriert die prädiktive Analyse sie, um Eintrittswahrscheinlichkeiten zu ermitteln.

Analysis models, such as regression analysis or decision trees, can be used to verify, falsify and improve the precision of prediction paths. The goal of this ongoing process is to uncover unknown correlations and to improve the response time for recognized scenarios or to make it possible in the first place.

The key, besides a solid database, is the combination of pattern recognition and evaluation, which can be supported by machine learning. In this way, event chains can be parameterized more precisely or unknown correlations can be uncovered.

Predictive analysis and machine learning

Quelle: Maimon, O. Z.; Rokach, L.: Data mining and knowledge discovery handbook. Springer, New York, 2010

In our TUP series KI we presented the different topics of machine learning. For the two scenarios of improving the prediction of known event chains, supervised learning is a promising approach that can improve the granularity and quality of prediction: The goal can be precisely defined and the associated conditions are known and described. Thus, in addition to data specialists, algorithms can also help to further extend the models, especially if many and complexly linked events are to be evaluated.

In order to discover unknown patterns in the first place, unattended machine learning can help if large, well-prepared data sets are available.

Benefits of predictive analysis in intralogistics

Besides the flow of goods, maintenance is a central point of predictive analysis. The article Concepts and methods for predictive maintenance in intralogistics by Michael Schadler, Norbert Hafner and Christian Landschützer shows that this form of analysis is a basic requirement for increasing efficiency through optimized maintenance cycles using the models Reliability Centered Maintenance, Total Productive Maintenance or Risk Based Maintenance.

Many companies still work here according to reactive measures or fixed maintenance intervals, which under certain circumstances can endanger ongoing operations.

The flow of goods is supported by improved forecasting, as more and smaller parameters are taken into account compared to traditional demand forecasts.

To the article overview
Back to Homepage