China
Africa
Explore chapters and articles related to this topic
Rule-Based Techniques
Published in Richard J. Roiger, Just Enough R!, 2020
In this chapter, we continue our discussion of supervised learning with a focus on rule-based machine learning techniques. The emphasis of Section 7.1 is on decision tree rules. In Section 7.2, we outline a fundamental covering rule algorithm. We then apply RWeka’s JRip covering algorithm to the churn data introduced in Chapter 6. In Section 7.3, we demonstrate an efficient technique for generating association rules. The Apriori(RWeka) association rule function is then utilized to find interesting relationships in a customer database of grocery store purchases. The focus of Section 7.4 is on Rattle, a graphical user interface (GUI) supporting many of the preprocessing, modeling and evaluation methods discussed throughout your text. We use Rattle’s interface to generate production rules with rpart,model customer churn with the randomForest function, and generate association rules with the apriori(arules) function.
Use of Association Rules for Cause-effects Relationships Analysis of Collision Accidents in the Yangtze River
Published in Adam Weintrit, Tomasz Neumann, Advances in Marine Navigation and Safety of Sea Transportation, 2019
B. Wu, J.H. Zhang, X.P. Yan, T.L. Yip
Association rule learning is a rule-based machine learning method for discovering hidden relationships between variables in a database from the perspective of data mining. When introducing it to ship collision accident analysis, after discovering the patterns of ship collision, it is meaningful to take countermeasures to cut off the necessary causation factors in an association rule. For example, an association rule for ship collisions in the Yangtze River is {accident area = anchorage} => {encounter scenarios = collision with stationary ship}. It shows that the collision with stationary ships will have a large probability to occur when ships are anchored at an anchorage. Therefore, the officer on watch should always take sharp lookout to prevent the occurrence of collision accidents when anchoring in the anchorage.
C
Published in Phillip A. Laplante, Dictionary of Computer Science, Engineering, and Technology, 2017
classifier system a rule-based machine learning system where a specific task is achieved through the interactive action of a set of classifiers. Each classifier is capable of accepting a set of specific messages which form its condition part, and will perform a specific action by generating a new message when its condition is satisfied. In addition to the classifiers, the system consists of a set of feature detectors which monitor the environment, a global message list which allows individual classifiers to observe and post messages, and a set of effectors for performing an action in response to the environment. Each message on the list is in the form of a binary string, and the condition and action of each classifier is specified as a symbol string of equal length over the alphabet {0,1, #}, where the “don’t care” symbol allows the classifier to accept more than one message type. To perform a specific task, the feature detectors obtain a set of measurements from the environment and represent these as messages on the global list. Each classifier then matches the messages with its own conditions, and on satisfaction generates new messages which are again posted on the list. External actions are performed by the system when messages on the global list directly actívate the effectors. See also learning classifier system.
Estimating energy savings of ultra-high-performance fibre-reinforced concrete facade panels at the early design stage of buildings using gradient boosting machines
Published in Advances in Building Energy Research, 2022
B. Abediniangerabi, A. Makhmalbaf, M. Shahandashti
In contrast to the forward approaches, data-driven approaches do not perform energy analysis based on physical principles or require detailed data about buildings. Data-driven techniques learn from historical data for prediction. In recent years, different statistical approaches, such as rule-based machine learning and statistical learning models, have been used by researchers to predict building energy uses. Statistical models provide this opportunity to effectively imitate BPSs while generating results much faster than BPSs (Singaravel et al., 2018). Rule-based machine learning algorithms are used in several studies for extracting hidden underlying structures and patterns from building energy use data (Ashouri et al., 2018). Such models are used as advisory systems where the outputs can be expressed as energy-saving recommendations. For example, Ashouri et al. (2018) used association rule mining to discover specific activity patterns for occupants that may reduce building energy uses. In another study, Abediniangerabi et al. (2020) proposed a rule-based framework to extract rules as recommendations for positive, negative, and neutral energy savings in favour of a new facade design over a baseline facade design. Rule-based machine learning algorithms provide several advantages, such as simplicity and interpretability of extracted rules. However, the extracted rules are qualitative, and they do not give the magnitude of energy savings regarding different design options. On the other hand, statistical learning models simply correlate building energy uses with explanatory variables. These models are generally classified into generalized linear models (GLMs), tree-based models, and artificial neural networks (ANNs) (Sutton et al., 2016).