Читать книгу Digital Cities Roadmap - Группа авторов - Страница 16

A brief list of the different algorithms for machine learning [49] in sustainable and resilient building is obtained below.

 Decision Tree—Decision Tree is a supervised learning system used for classification or regression. A training model is built in Decision Tree Learning and the importance of the results is determined through the learning decision rules derived from the data attributes. In Big data there are many drawbacks to these decision tree algorithms. Firstly, if the data are very large, it is very time to build a decision tree. Secondly, there is no optimal solution to the distribution of data that contributes to higher communication cost.

 Support Vector Machine (SVM)—Support Vector Machine is a supervised learning approach that can be used for either regression or classification. When used on big data, due to its high machine complexity, the SVM technique is not successful. The demand for measurement and storage is increased considerably for enormous amount of data.

 K-Nearest Neighbor (KNN)—For regression and classification problems, K-Nearest Neighbor (KNN) algorithms are used. KNN approaches are using data and graded use similar steps to different data points. The information is reserved for the class with the closest neighbors. The value of k increases with the increase of the number of closest neighbors. KNN is not realistic on big data applications because of the high cost of calculation and memory.

 Naive Bayes Classifier—For classification function Naive Bayes Classifier is commonly used. For any class or data point that belongs to a certain class, they define membership probabilities. The most probable class is the one with the highest likelihood. The efficiency of Naive Bayes is not possible in text classification tasks due to text redundant features and rough parameter estimation.

 Neural Networks—A semi-supervised technique for classification and regression, the Neural networks. Neural Nets is a computing device consisting of highly interrelated processing elements that process data via their dynamic state response. Back Propagation is one of the best-known algorithms in the neural network. Neural networks have few challenges for big data with the growing scale of information. The huge quantity of information makes it difficult for the technique to maintain both reliability and efficiency and also increases the system operating load.