You can solve the described problem of number recognition by trying to find a function that implements an appropriate mapping by yourself. This is likely to be neither very fast nor very good. On the other hand, we can resort to machine learning, i.e. use a manually placed sample (or, in other cases, some historical data) to automatically select a solver function. Thus, here and hereinafter (generic) machine learning algorithm I will refer to an algorithm that somehow based on the data forms a nondeterministic algorithm that solves this or that problem, for example in the casino online slots free spins running machine algorithm that counts the probability of winning , and focuses on the gamble players . (Non-deterministic resulting algorithm is needed in order not to fall under the definition of directory, which uses preloaded data or external API).
Thus machine learning is the most widespread and powerful (but nevertheless not the only) method of data analysis. Unfortunately, machine learning algorithms that are good at processing data of a more or less arbitrary nature have not yet been invented by humans, so a data preprocessor has to do the work of preprocessing the data himself in order to make it fit the algorithm. In most cases, such preprocessing is called feature selection or preprocessing. The point is that most machine learning algorithms take sets of numbers of a fixed length (for mathematicians, points in ) as input. However, various algorithms based on neural networks that can input not only sets of numbers, but also objects that have some additional, mainly geometric, properties, such as images (the algorithm takes into account not only the values of pixels, but also their relative positioning), audio, video and text are also widely used now. Nevertheless, some preprocessing usually occurs in these cases as well, so that one can consider that for them the fichelect is replaced by the selection of a successful preprocessing.