In line with the strategy of our university of applied sciences, we work with scientific partners to support companies in the generation of added value using process models and methods of artificial intelligence.
FITAPP is an Austrian start-up company founded in 2016 which provides a complete fitness platform. One of the main aspects of the FITAPP platform is the social component, which allows users to interact and motivate one another. As the European market for fitness applications is largely saturated, FITAPP managed to gain significant market shares in Brazil in particular.
The main problem the company faces is the small number of paid subscriptions compared with the number of downloads. What is more, only a tiny fraction of users (about three percent) run the app 30 days after downloading it. A third problem is the large number of unregistered users, for whom no data are stored.
The long-term goal of the project is to increase the number of paid subscriptions. The analysis of the existing data therefore aims to find useful information about customer behaviour, which can then be used to come up with specific marketing strategies.
Image recognition using neural networks for Landwirt.com
More than 1,600 dealers from across Europe offer over 70,000 agricultural machines and tractors on Landwirt.com, one of the leading Austrian internet portals with over two million unique clients.
The aim of the project is to make it easier for Landwirt.com customers to create advertisements. An artificial neural network will be developed to classify the uploaded images as precisely as possible by machine type, make and perhaps even model. This classification can then be used to automatically generate key components of the ad and make them available to the customer. The image recognition feature will also detect ads that have been entered in the wrong category. This not only helps users, it can also potentially prevent fraud; for example, machines whose sale is subject to a fee can sometimes be found in free ad categories.
The project is scientifically supported by experts from Know-Center GmbH, who helped create the CNNs (convolutional neural networks).
Forecasts of the population development of the city of Graz
This pilot project aims to investigate and compare different methods for forecasting population development with special consideration of the number of pupils. The forecast is broken down by age and other parameters (such as gender), at least at district level. A key question is to what extent a fine-grained analysis on the basis of statistical enumeration districts is possible and reasonable.
The data sources are the local register of residents and records of school enrolment and afternoon child care; the data is pseudonymised and all other aspects of data protection are also strictly observed. The analyses are carried out using standard methods of multivariate statistics as well as approaches from the fields of system dynamics, i.e. modelling with differential equations, possibly agent-based approaches and always machine learning methods such as artificial neural networks.
The aim of the research is to compare the different approaches and to identify the most suitable method for making forecasts for a timescale of twenty years. The implementation of this method, which is not part of the pilot project, should enable statistics experts not only to generate forecasts but also to update them regularly using new data. Such dynamic forecasts can help the City of Graz to plan infrastructure measures in a targeted manner.
The pilot project of the Austrian Federal Railways (ÖBB) has been set up to carry out a wheelset analysis of different Railjet train sets. The first part of the project involved the cleaning of factory and service data, which were then extended to perform an analysis of the wheelset measurements. To help understand the data and to identify connections and correlations, the data were presented graphically. This involved the following development steps: sample plots, aggregate plots and pair plots.
The next step was modelling and evaluation. The multiple regression method was used to model the wear and tear of the wheelsets of the Railjet trains. Since the manual analysis of such a large number of models would be impractical and error-prone, two tools for automated analysis were additionally developed in Python. One of the tools splits the set of models into two subsets and generates several aggregated forecast quality indicators from the MSE of the individual models. The other tool ranks all models based on the identified R2 values and automatically generates an evaluation report in table form.
Real-time logistics optimisation for disposal at construction sites
The pilot project carried out with pink:robin gmbh, Saubermacher Dienstleistungs AG, looks into methods for optimising complex logistics tasks. The methods are designed to serve as a basis for further automating the construction site disposal concept that is currently in use. The project is specifically about improving the planning and execution of skip and container orders. These orders are currently planned and assigned to the right truck types manually by Wastebox partners on a daily basis, which is a complex and difficult task. The generated order list is then processed by the truck drivers.
The project aims to create optimal processing sequences to enable better, and more cost-efficient and ecologically sound route planning. In order to implement this adequately, one must take into account a large number of ancillary conditions regarding types of order, types of disposal, skip combinations and vehicle types. In addition, the choice of recycling stations, which may depend on the above-mentioned categories, must be optimally adapted to the routes.
An essential aspect of the project is the implementation of dynamic order assignment. This dynamic perspective involves including new orders into the existing route by checking same-day processing. The resulting new order sequence is synchronised with the system and transmitted to the drivers. In addition, the optimal routes depend on the traffic situation, which is why the routes have to be regularly checked and adapted, where necessary.
Data analysis for Theaterholding Graz / Steiermark GmbH
The aim of the project is to detect surprising interconnections in the event-specific data of Theaterholding. Can customers be grouped into clusters according to their preferences? Do these clusters perhaps reflect unexpected genre combinations? Is there a link between time of purchase and distance to the venue? Are there striking patterns for the different genres when it comes to the sales data for the relevant events?
The FIT4BA pilot project will use the methods of classical statistics and heuristic data analysis to find answers to these and similar questions. The Institute of Statistics at Graz University of Technology provides scientific support in the implementation of this exciting data analysis project. Finally, a software tool, which will be developed together with Theaterholding, will facilitate the daily work of the company by providing informative analyses.
The aim of the project is to search for interesting patterns in the ERP data of the different Brau Union sites in Austria and to generate important findings for controlling and quality management. The project adapts current statistical methods to suit the requirements of Brau Union using the R software package. These will also be provided to Brau Union as a software tool for optimising their daily work. Scientific support for the project is provided by the Institute of Applied Statistics at JKU Linz.
As part of this pilot project, the FH JOANNEUM team together with the Data Science Group of RLB Styria, investigates methods and strategies from the fields of big data and artificial intelligence that can benefit business activities in the banking sector. The project focuses on transparent, explainable methods, while also looking into complementary approaches.
Strict attention is paid to data protection, and the ethical aspects of the project have undergone intense scrutiny. Binding guidelines were defined to ensure that big data methods are never used to customers' disadvantage, and specifically that they do not lead to inequality or discrimination. The research focuses on methods for identifying cross- and up-selling potential. Customers' preferences and their affinity for specific products is of particular interest in this context. The project is also about the suitability of interaction channels and identifying effective contact frequencies.
The precise assessment of these values can lead to more targeted marketing measures later on. This will enable RLB to make efficient use of its marketing capacities. Customers are provided with information about products that are likely to be relevant to them in the most agreeable way possible. In a planned follow-up project, a promising method will be adapted to the needs of the sales department and tested.