Machine Learning is giving computers the ability to automatically learn and improve from experience without human intervention and without being explicitly programmed.
Machine Learning tasks are typically classified into three broad categories:
Supervised learning: the computer is presented with example inputs and their corresponding outputs, the goal being to learn a general rule that maps inputs to outputs. This can be used in operational applications, for instance assessing the potential of a new client or the risk of a loan applicant. A distinctive feature is that the computer will keep learning as it gets more experience (i.e. more client history).
Unsupervised learning: no labels are given to the learning algorithm, leaving it on its own to find structure in its input. This can be used as a goal in itself (discovering hidden patterns in data, anomaly detection) or a means towards an end (data mining, exploratory data analysis as a first step to feature learning).
Reinforcement learning is less relevant to Business Intelligence: a computer is provided feedback in terms of rewards and punishments as it navigates in a dynamic environment in which it must perform a certain goal, such as driving a car or playing a game again an opponent.
Another way to understand what is Machine Learning is to consider the output expected by the ML system:
Classification, where inputs are divided into two or more classes, and the program must produce a model that assigns unseen inputs to one or more (multi-label classification) of these classes. Typical examples would be customer and product segmentations (in the "supervised" category if the classes are known beforehand).
Regression is also a supervised problem, where the outputs are continuous rather than discrete. Typical applications would be predictive analysis (sales forecast, estimation of optimum pricing for instance)
Clustering, where set of inputs is to be divided into groups. Contrary to classification, the groups are not known beforehand, making this an unsupervised task.
Dimensionality reduction, mapping input into a lower-dimensional space, either as goal in itself (for visualisation or identifying key drivers) or a first step before another operation, typically clustering.
Business Intelligence analysts have been using statistical and probabilistic algorithms for years, but the technological progress represented by Machine Learning has dramatically expanded the realm of analyses and potential ROIs.
Typical applications include:
Given its broad scope, any industry can benefit from Machine Learning business insights, provided a critical mass of data is available.
Although Machine Learning and Big Data have emerged around the same time, Machine Learning is not restricted to Big Data or online implementations.
Even industries not data-intensive by nature will derive significant benefits from Machine Learning, simply by exploiting existing data with modern techniques.
Whereas Machine Learning is part of Data Science, Big Data can be defined as data sets that are so large or complex that traditional data processing application software is inadequate to deal with them, or described by its characteristics in terms of Volume, Variety, Velocity, Variability and Veracity.
Regardless of the definition used, the existence of Big Data is a direct consequence of the growth of the Internet and of smartphone availability.
Even though the two concepts are not directly related, Machine Learning can prove more effective at handling Big Data than traditional statistical techniques, taking advantage of continuous learning processes as described above and/or stochastic or Bayesian approaches.
But conversely, the applicability of Machine Learning is not restricted to very large datasets. Applying Machine Learning algorithms to offline medium-sized datasets is likely to provide meaningful business insights, that can be further improved through data-enrichment techniques.
Initially part of Artificial Intelligence, Machine Learning started as a separate field in the 1990s when it set its goal to tackling solvable problems of a practical nature, as opposed to achieving artificial intelligence.
It shifted focus away from the symbolic approaches it had inherited from AI, and toward methods and models borrowed from statistics and probability theory. Machine Learning's development has been exponential since, facilitated by the increasing availability of digitised information, of cheaper and more powerful computational processing, and of affordable data storage.
This has led Machine Learning to become increasingly visible, either through ubiquitous applications (such as email filtering, web search engines, automated translation, online retailers' recommendations, face recognition at airports and in digital cameras…) or headline grabbing projects (beating World Champions of Chess and Go, going towards self-driving vehicles…).
The same underlying factors, together with the introduction of innovative and powerful algorithms (such as neural networks or random forests), the use of Bayesian statistics (where hypotheses are modelled as statistical variable themselves) have given new tools for Business Intelligence analysts, either to exploit new data resulting from online developments, or re-visit existing data with enhanced models.
Copyright © 2017 B3 METRIX Ltd- All Rights Reserved.