A similarity measure to assess the stability of classification trees

Briand, B.; Ducharme, Gilles R.; Parache, V.; Mercat-Rommens, C.

doi:10.1016/j.csda.2008.10.033

Cited by 28 publications

(22 citation statements)

References 18 publications

Supporting

Mentioning

Contrasting

Unclassified

Order By: Relevance

“…Analysing the results in detail, we found that the top-scored classifiers among the 126 tested coincide with those showing good performance in traditional machine learning works [36,38]. In our case, RF reaches the highest accuracy in validation.…”

Section: Discussionsupporting

confidence: 59%

Detecting and Classifying Human Touches in a Social Robot Through Acoustic Sensing and Machine Learning

Alonso-Martín

Gamboa-Montero

Castillo

et al. 2017

Sensors

View full text Add to dashboard Cite

An important aspect in Human–Robot Interaction is responding to different kinds of touch stimuli. To date, several technologies have been explored to determine how a touch is perceived by a social robot, usually placing a large number of sensors throughout the robot’s shell. In this work, we introduce a novel approach, where the audio acquired from contact microphones located in the robot’s shell is processed using machine learning techniques to distinguish between different types of touches. The system is able to determine when the robot is touched (touch detection), and to ascertain the kind of touch performed among a set of possibilities: stroke, tap, slap, and tickle (touch classification). This proposal is cost-effective since just a few microphones are able to cover the whole robot’s shell since a single microphone is enough to cover each solid part of the robot. Besides, it is easy to install and configure as it just requires a contact surface to attach the microphone to the robot’s shell and plug it into the robot’s computer. Results show the high accuracy scores in touch gesture recognition. The testing phase revealed that Logistic Model Trees achieved the best performance, with an F-score of 0.81. The dataset was built with information from 25 participants performing a total of 1981 touch gestures.

show abstract

Section: Discussionsupporting

confidence: 59%

Detecting and Classifying Human Touches in a Social Robot Through Acoustic Sensing and Machine Learning

Alonso-Martín

Gamboa-Montero

Castillo

et al. 2017

Sensors

View full text Add to dashboard Cite

show abstract

“…Its integrated variable importance measure provides a ranking of the variables according to their explanatory power (see Calle and Urrea, 2011). In fact, the RF has been stated to be the best classifier in comparison to a total number of 179 classifiers tested for “Average Accuracy” and “Friedman Ranking” (Fernández-Delgado et al, 2014). …”

Section: Methodsmentioning

confidence: 99%

Importance of Ecological Variables in Explaining Population Dynamics of Three Important Pine Pest Insects

Hentschel

Möller²,

Wenning³

et al. 2018

Front. Plant Sci.

View full text Add to dashboard Cite

Climate change challenges forest vitality both directly by increasing drought and heat periods and indirectly, e.g., by creating favorable conditions for mass outbreaks of phyllophagous insects. The large forests dominated by Scots pine (Pinus sylvestris L.) that cover the lowland regions in northeast Germany have already been affected regularly by cyclic mass propagations of defoliating insect species in the past with climate projections implying an even more advantageous environment for devastating outbreaks in the future. To improve predictive and responsive capacities we have investigated a wide range of ecological parameters to identify those most strongly related to past outbreak waves of three central species. In total, we analyzed 3,748 variables covering stand and neighborhood properties, site quality, and climatic conditions for an area of roughly 750,000 ha of pine forests in the period 2002–2016. To reflect sensitivity against varying climate, we computed “floating windows” in relation to critical phenological phases of the respective insects. The parameters with the highest explanatory power resulted from the variable importance measures of the Random Forest (RF) methodology and have been evaluated by a 10-fold cross-validation process. Our findings closely reflect the known specific gradation patterns and show that relative variable importance varies with species. While Lymantria monacha L. feeding was mainly dependent on the surroundings of the respective stand, Diprion pini L. proved to be almost exclusively susceptible to climatic effects in its population dynamics. Dendrolimus pini L. exhibited a mixed pattern of variable importance involving both climatic and forest structure parameters. In many cases the obtained statistical results support well-known ecological cause-effect relations and long-term population change dynamics. The RF delivered very high levels of sensitivity and specificity in the developed classifications and proved to be an excellent tool to handle the large amounts of data utilized for this study. While the presented classification approach may already support a better prognosis of the amplitude during the outbreak culmination, the obtained (most important) variables are proposed as preferable covariates for modeling population dynamics of the investigated insect species.

show abstract

“…Two key insights have driven the effectiveness of random forests models for complex tasks while avoiding overfitting [8,9] The field of deep learning has exhibited success in developing models for tasks ranging from image processing [29,30,31,32,33,34,35] to text generation [36,37,38] and games [39]. Several reviews of the field give a summary of recent breakthroughs and developments [40,41,42,12,43].…”

Section: Machine Learning Algorithmsmentioning

confidence: 99%

“…More broadly, traditional ML methods include techniques such as linear and polynomial regression, k-nearest neighbors, support vector machines, Gaussian processes, and random forests. Of these, we focus only on the latter because they have demonstrated widespread success for complex modeling applications [8,9]. Random forests are based on an ensemble of decision trees, where decisions are based on the model parameters to provide estimates of the target.…”

Section: Introductionmentioning

confidence: 99%

Deep learning for presumed probability density function models

Frahan

Yellapantula

King

et al. 2019

Combustion and Flame

View full text Add to dashboard Cite

In this work, we use machine learning (ML) techniques to develop presumed probability density function (PDF) models for large eddy simulations of reacting flows. The joint sub-filter PDF of mixture fraction and progress variable is modeled using various ML algorithms and commonly used analytical models. The ML algorithms evaluated in the work are representative of three major classes of ML techniques: traditional ensemble methods (random forests), deep learning (deep neural networks), and generative learning (conditional variational autoencoder (CVAE)). The first two algorithms are supervised learning algorithms, and the third is an unsupervised learning algorithm. Data from direct numerical simulation of the low-swirl burner [1] are used to develop training data for sub-filter PDF models. Models are evaluated on predictions of the sub-filter PDFs as well as predictions of the filtered reaction rate of the progress variable, computed through an integral of the product of the sub-filter PDF and the conditional means of the reaction rate. This a-priori modeling study demonstrates that deep learning models for presumed PDF modeling are * Corresponding author Email addresses: marc.henrydefrahan@nrel.gov (M. T. Henry de Frahan), ray.grout@nrel.gov (R. W. Grout) three times more accurate than analytical β-β PDF models and linear regression models. These models are as accurate as random forest models while using five times fewer trainable parameters and being 25 times faster for inference. In this work, conditional unsupervised learning did not present additional advantages beyond supervised learning with a feed-forward neural network. We illustrate how models generalize to other regions of the flow and develop criteria based on the Jensen-Shannon divergence to quantify the performance of a model on new data.

show abstract

A similarity measure to assess the stability of classification trees

Cited by 28 publications

References 18 publications

Detecting and Classifying Human Touches in a Social Robot Through Acoustic Sensing and Machine Learning

Detecting and Classifying Human Touches in a Social Robot Through Acoustic Sensing and Machine Learning

Importance of Ecological Variables in Explaining Population Dynamics of Three Important Pine Pest Insects

Deep learning for presumed probability density function models

Contact Info

Product

Resources

About