Читать книгу Digital Cities Roadmap - Группа авторов - Страница 47

Massive information generated by detectors, wearables and other technologies have comprehensive knowledge regarding the product background and construction status, which can be used to construct SB management.

The ML algorithms can be split between walking and jogging lessons between potential data points. Without any human intervention, it is fairly straightforward for ML to build sophisticated software systems. They are applicable in SB environments to many real-life issues. Self-learning and collaboration frameworks may also be built and created. ML algorithms can research and render input data predictions.

The furnace of both the nesting is an illustration of a device that, like the resident desired, maintains a different climate in a certain location and at those periods of the day. These are applications like the Amazon’s Alexa that can understand from words, whereas some learn from even more nuanced behaviors. In order to create intelligent systems that can sense and respond according to contextual shifts, ML strategies have been used extensively.

The four main categories of Machine Learning are: Supervised Learning, Semi-Supervised learning, Unsupervised Learning and Reinforcement Learning. Figure 1.23 shows ML styles. The next explanation of these groups is the integration of this methodology in Table 1.5.

Supervised Learning: The ML model is built through an inputtraining cycle which continues until the model achieves the required accuracy. Some of the examples of commonly monitored ML algorithms include: Naive Bayes model, decision-making tree, linear discriminatory functions (SVMs), hidden Markov models (HMMs), instance education (for instance, k neighbor learning), ensembles (bagging, boosting, random forest), logistic regression, genetic algorithms and so on. ML algorithms are also common to use as an alternative algorithm. Monitored methods of learning are commonly used to solve different problems in SB.

Classification: Classification algorithms are intended to classify an instance into specific discreet categories. Due to two data sets (labeled and unscheduled data sets), the labeled data set is used to train.

Figure 1.23 Machine learning techniques.

Decision Tree Algorithms: The decision tree approach is a key predictive ML modeling approach, which constructs a decision model based on the real values of the data’s features. For both classification and regression issues, decision trees can be used.

Bayesian Algorithms: For classification and regression questions, the Bayesian approaches use Bayes’ theorem. Naive Bayes, Naive Gauss, Bayesian beliefs network, Bayesian Network and Bayesian Network are most general.

Support Vector Machine (SVM): SVM is one of the most commonly used for a large range of problems of statistical learning, including the identification of the face and object, classification of messages, spam-related detection and handwriting analysis.

Artificial Neural Network Algorithms (ANNs): The mechanism of the biologic neural networks inspires the ANN models. For regression and classification problems, ANN models are frequently used. The major algorithms are: perceptron, back-propagation (back-propagation), Hopefield network and radial feature network (RBFN).

Deep Learning Algorithms: Deep learning techniques reflect a type of advanced NANs in which deep (many layers consisting of several linear and non-linear transformations) architecture is used.

Hidden Markov Models (HMM): An HMM is a twice stochastic cycle with a secret corresponding stochastic system that can be found in the series of symbols that another stochastic mechanism generated.

Statistical Analysis: A critical path is a collection of scenarios; sets typically contain high dimensionality, a wide range of cases and continuous changes.

Table 1.5 Difference between of ML techniques.

Category	Type	Algorithms	Pros	Cons	Applicability in SBs
Supervised Learning	Classification	Neural networks	Request little statistical training: Can detect complex non-linear relationships	Computational burden; Prone to Overfitting; Picking the correct topology is difficult; Training can take a lot of data	Used for classification control and automated home, appliances, next step/action prediction
		SVM	Can avoid overfitting using the regularization; expert knowledge using appropriate kernels	Computationally expensive; Slow: Choice of kernel models and parameter, sensitive to overfitting	Classification and regression problems in SBs such as activity recognitions, human tracking, energy efficiency services
		Bayesian networks	Very simple representation does not allow for rich hypothesis	You should train a loge training set to me it well	Energy management and human activity recognition
		Decision trees	Non-parametric algorithm that it easy to interpret and explain	Can easily overfit	Patient monitoring, healthcare services, awareness and notification service
		Hidden Markov	Flexible generalization of sequence profiles; can handle	Requires training using annotated data: Many unstructured parameters	Daily living activities recognition classification
		Deep Learning	Enables learning of feature rather than hand tuning: Reduce the need for feature engineering	Requires a very large amount of labeled data. computationally really expensive, and extremely hard to tune	Modeling occupied a behavior, and in human voice recognition and monitoring systems; Context-aware SB services
	Regression	Orthogonal matching pursuit	Fast	Can go seriously wrong if there are severe outliers or influential cases	For regression problem such as energy efficiency services in SBs
	clustered based	Straightforward to understand and explain, and can be regularized to avoid overfitting	It is not flexible enough to capture complex patterns	Gesture recognition
	Ensemble methods	N/A	Increased model accuracy through averaging as the number of model increases	Difficulties in interpreting decisions; Large computational requirements	Human activity recognition and energy efficiency services
	Time Series	N/A	Can model temporal relationships; Applicable to settings where traditional between subject design are impossible or difficult to implement	Model identification is difficult; Traditional measures may be inappropriate for TS designs; Generalizability cannot be inferred from a single study	Occupant comfort services and energy efficiency services in SBs
Unsupervised learning	Clustering	KNN	Simplicity: Easy to implement and interpret; Fast and computationally efficient	High computation cost; lazy learner	Human activity recognition.
K-pattern clustering	Simple; Easy to implement and interpret; Fast and computationally efficient	Only locally optimal and sensitive to initial points; Difficult to predict K-Value	Predict user activities in smart environments
Others	N/A	N/A	N/A
Semi- Supervised Learning	N/A	N/A	Overcome the problem of supervised learning having not enough labeled data	False labeling problems and incapable of utilizing out-ofdomain samples	Provide context aware services such as health monitoring and elderly care services
Reinforcement learning	N/A	N/A	Used “deeper” knowledge about domain	Must have a model of environment; must know where actions lead in order to evaluate actions	Lighting control services and learning the occupants, preferences of music and lighting services.

Ensemble Methods: The community of classification models that are educated separately and then predictions are merged in a way that generates the ultimate prevision, also referred to as classifier ensemble.

Unsupervised learning implies designing algorithms to use data that have no labeling to evaluate the behavior or structure being analyzed. The algorithm is the best techniques to work on its own to discover patterns and information that was previously undetected. Clustering, anomaly, detection, Neural Networks etc. are all the examples of unsupervised learning.

Clustering: The internal groupings in the products, such as the grouping of customers, are investigated through a clustering problem. Modeling approaches including centroid-based and hierarchical are typically organized through clustering techniques.

Association: The question of the association rule is used to classify laws that describe significant quantities of input data, such as individuals who purchase X products, who also purchase Y objects. Association research can be achieved by evaluating rules for repeated if/then statement inputs and utilizing help requirements and trust to distinguish associations between unconnected data in a relational database.

Semi-Supervised Learning: Semi-controlled instruction is between approaches regulated and unregulated. Information is a labeled and blank experimental combination. Such architectures are synthetic are intended to consider and counteract the weaknesses of the main groups.

Reinforcement Learning: To order to optimize the principle of accrual compensation, enhanced learning, an ML area influenced by behavioral science, is concerned with the way virtual agents are to work to an environment. RL algorithms are used to learn policy of control, particularly if no prior information exists and a large amount of training data are available.