8 – 10 Classification, on the other hand, entails predicting to which category an example belongs. 8 – 10 Regression involves predicting numeric data, such as test scores, laboratory values, or prices of an item, much like the housing price example. The most common supervised learning tasks are regression and classification. This concept will be discussed further in the section on performance evaluation. Insofar as the validation set may differ from the test set, the performance of the algorithm may or may not generalize. In this way, the algorithm is tuned by the validation set. In each iteration, the performance of the algorithm on the training data is compared with the performance on the validation dataset. 8, 9 The basic steps of supervised machine learning are (1) acquire a dataset and split it into separate training, validation, and test datasets (2) use the training and validation datasets to inform a model of the relationship between features and target and (3) evaluate the model via the test dataset to determine how well it predicts housing prices for unseen instances. The performance of the algorithm is evaluated on the test dataset, data that the algorithm has never seen before. 8 – 10 That is, features, x, are mapped to the target, Y, by learning the mapping function, f, so that future housing prices may be approximated using the algorithm Y = f( x). This approach is supervised because the model infers an algorithm from feature-target pairs and is informed, by the target, whether it has predicted correctly. 8, 9 Supervised learning uses patterns in the training dataset to map features to the target so that an algorithm can make housing price predictions on future datasets. 8, 9, 11 Datasets are generally split into training, validation, and testing datasets (models will always perform optimally on the data they are trained on). 8, 9, 11 The target is the feature to be predicted, in this case the housing price.
Features are the recorded properties of a house that might be useful for predicting prices (e.g., total square-footage, number of floors, the presence of a yard). 8, 9, 11 Each instance represents a singular observation of a house and associated features. To begin, the company would first gather a dataset that contains many instances. Suppose the real estate company would like to predict the price of a house based on specific features of the house.
This review summarizes machine learning and deep learning methodology for the audience without an extensive technical computer programming background. These more complicated tasks are where ML and DL methods perform well. Although symbolic AI is proficient at solving clearly defined logical problems, it often fails for tasks that require higher-level pattern recognition, such as speech recognition or image classification. For example, if one were to program an algorithm to modulate room temperature of an office, he or she likely already know what temperatures are comfortable for humans to work in and would program the room to cool if temperatures rise above a specific threshold and heat if they drop below a lower threshold. These rules, written by humans, come from a priori knowledge of the particular subject and task to be completed. However, AI includes approaches that do not involve any form of “learning.” For instance, the subfield known as symbolic AI focuses on hardcoding (i.e., explicitly writing) rules for every possible scenario in a particular domain of interest. That is, they are within the realm of AI ( Fig. 1). In 1956, a group of computer scientists proposed that computers could be programmed to think and reason, “that every aspect of learning or any other feature of intelligence, in principle, be so precisely described that a machine be made to simulate it.” 7 They described this principle as “artificial intelligence.” 7 Simply put, AI is a field focused on automating intellectual tasks normally performed by humans, and ML and DL are specific methods of achieving this goal.