Machine Learning - a look behind the hype
The term Machine Learning is often used for marketing and sales purposes. It should therefore not be forgotten that behind the buzzword, there is in fact real added value for companies. One of my preferred lecture topics is to explain, in a comprehensible way, what happens conceptually in different Machine Learning use cases and what the applications really look like in practice. In a single blog post, the topic is of course not dealt with in full depth. Therefore, I would like to use my first contribution to the topic to give an introduction to the different methods of Machine Learning - and some first insights into its use in practice.
There are many different definitions of Machine Learning - one that I like is: Machine Learning happens when the computer learns to make decisions on its own from data. There are different ways to derive these decisions, or algorithm classes. Examples are supervised, unsupervised, partially supervised and reinforcement learning or genetic algorithms. In this paper, we discuss the two most common types of supervised and unsupervised learning. I will discuss other methods, including their advantages and disadvantages, in later articles.
Machine Learning Types and their Application in Practice
The supervised learning method is a guided technique in which training data including previously observed target values are provided for the construction of the model. The data should be independently assigned to given classes or groupings. One practical example comes from the fraud detection sector, where software can use Machine Learning to quickly identify patterns and anomalies that indicate fraud (the classes are therefore fraud/no fraud). If, for example, a credit card withdrawal takes place in Germany and ten minutes later with the same card in Thailand, it must be fraud. A major challenge lies in the fact that fraud patterns are constantly changing. In accordance to this, the Machine Learning algorithms must learn continuously.
Another method of Machine Learning is unsupervised learning. In contrast to supervised learning, this method lacks known target values. The machine independently tries to recognize patterns or statistical peculiarities in the input values. A good example of how it can be applied is its use in chemical parks. Companies in a chemical park pay for the use of its infrastructure, for example in the form of steam or electricity, among other things. Chemical parks also discharge toxic waste water for the companies. Each company has a sewer that leads into the main sewer system, which in turn leads the waste water into a sewage treatment plant. The operator of the chemical park would need to know whether all the companies’ waste water discharge systems are running according to plan, as otherwise the operator could, among other things, suffer financial damage as a result of the shutdown of the sewage treatment plant and, consequently, of the chemical park. In addition to information on companies - such as water, electricity or steam collection, continuous automated sampling also provides information on how much waste water is currently being pumped into the sewer system, what the temperature is, and so on. Using these values, anomalies can be directly detected over time with the help of Machine Learning.
Segmentation using Machine Learning
Another topic that is of great importance for many industries and can be optimized with the help of Machine Learning is data segmentation. For example, clustering algorithms can help companies identify different customer segments. Retail companies can use customer information, product information, information about current promotions and much more to identify different types of customers and gain important insights into their target group structure. They learn what unites or distinguishes the various segments. Thus, targeted marketing measures are possible according to these target groups.
Clustering can also be used in machine and systems engineering. I know an example of a machine engineer who produces thousands of different tools for CNC milling, but did not catalogue them properly, e.g. by type. However, a lot of information about these tools, such as material strength or geometric properties, was already stored in a database. In order to find out which tool groups exist and what distinguishes them, he was able to cluster this data in a short period of time with the help of Machine Learning and not only increase the master data quality, but also gain an unprecedented overview or insight into his portfolio.
The assumption that you need very high data quality in order to be able to use Machine Learning algorithms at all successfully is not entirely correct. You can often improve data quality using Machine Learning.
There are numerous examples of how Machine Learning can help companies optimize processes and data. Behind the buzzword there is real added value! Thanks to supervised and unsupervised Learning methods, business processes can be automated, insights into processes can be obtained and decisions can be made data-driven.
In my next articles, I will tell you more about Artificial Intelligence in practice and discuss other topics such as predictive maintenance and data quality in detail.
Do you already use Artificial Intelligence in your company? Are you looking for further information on Machine Learning in practice? Contact me.