Machine Learning

Machine learning is a specialized area of statistics and artificial intelligence which focuses on identifying trends within data. Machine learning is often implemented with big data, where there may be millions or billions of data points or where each data point may be represented by hundreds or thousands of characteristics. In such situations, traditional statistical approaches may be limited by the extensive nature of the data. By making use of existing data, machine learning algorithms can update themselves over time so that the resulting models “learn” as more data are collected.

There are numerous machine learning algorithms, and the families of algorithms are often grouped together based on whether they are supervised or unsupervised algorithms. Supervised machine learning algorithms identify trends in the data and then use knowledge of the true classification or the true value to select the model with optimal prediction accuracy. Within supervised machine learning algorithms, classification algorithms are used to predict the nominal category for each case and regression algorithms are used to predict the numerical value for each case. Unsupervised machine learning algorithms are often implemented when knowledge of the true classification for each case cannot be known. Consequently, unsupervised machine learning algorithms cluster cases based on the available features for each case.

Regardless of whether machine learning algorithms are supervised or unsupervised, the primary role of machine learning models is to generate predictions about future cases using the existing data, meaning the accuracy of those predictions is paramount. In supervised algorithms, performance indices are used to evaluate the effectiveness of the resulting machine learning model. In unsupervised algorithms, evaluating model performance is more difficult since the true classification of the cases is not known.

Because machine learning is a complex and abstract topic, a hypothetical data set is first be presented as a running example before defining the necessary terms and explaining the inner workings of machine learning.