Machine learning and artificial intelligence have been the hottest technological trends during the last decade. As always, when the hype was highest the expectations also were that it solves all problems like a silver bullet. There were hopes but also fears that machine learning will change almost everything. After big corporations, and smaller ones, started up and finished machine learning projects, people noticed that machine learning also has its own place and time. But what is machine learning?
In short, the use of machine learning methods is an entity with three significant components:
- mathematic algorithms implemented as code
- digitally processable data
- an IT infrastructure with apps and services that make it possible to train the model and use the results
Machine learning algorithms, both the more traditional ones and newer methods created by recent research, can readily be found implemented in different libraries and frameworks. Understanding is required for utilising the methods, so that it is also possible to assess the suitability of the method.
However, one does not need to start developing the machine learning algorithms from scratch. It is crucial to understand what algorithm will give the maximum benefit with what kind of a problem.
The data have proven to be the most laborious part in setting up machine learning projects, and from the methodological point of view, they are in the very centre. First, there has to be enough data, which as a coarse estimate means at least thousands of observations of the figure one wants to model.
Secondly, the data must be of sufficient quality. Thorough secondary processing secures the quality of data and takes into account the chosen machine learning method. Defining the data, i.e. the variables used in the prediction, and the predicted subject itself, often requires significant cooperation between the different business sectors.
The utilisation of machine learning methods in the daily service business is in the end just data-driven application development governed by the same rules as any application development project. The additional demands come through data quality, data use and data capture.
Utilising data through machine learning methods starts with understanding the possibilities. We have had good experience from making half-day low-threshold machine learning workshops, certain digital workshops, where the subject is presented and the focus is on finding in your own business problems that are solvable by machine learning.