Random Forest Ensemble Research Articles

PSuc-Lys: Predict lysine succinylation sites in proteins with PseAAC and ensemble random forest approach

Being one type of post-translational modifications (PTMs), protein lysine succinylation is important in regulating varieties of biological processes. It is also involved with some diseases, however. Consequently, from the angles of both basic research and drug development, we are facing a challenging problem: for an uncharacterized protein sequence having many Lys residues therein, which ones can be succinylated, and which ones cannot? To address this problem, we have developed a predictor called pSuc-Lys through (1) incorporating the sequence-coupled information into the general pseudo amino acid composition, (2) balancing out skewed training dataset by random sampling, and (3) constructing an ensemble predictor by fusing a series of individual random forest classifiers. Rigorous cross-validations indicated that it remarkably outperformed the existing methods. A user-friendly web-server for pSuc-Lys has been established at http://www.jci-bioinfo.cn/pSuc-Lys, by which users can easily obtain their desired results without the need to go through the complicated mathematical equations involved. It has not escaped our notice that the formulation and approach presented here can also be used to analyze many other problems in computational proteomics.

Lymph Diseases Prediction Using Random Forest and Particle Swarm Optimization

This research aims to develop a model to enhance lymphatic diseases diagnosis by the use of random forest ensemble machine-learning method trained with a simple sampling scheme. This study has been carried out in two major phases: feature selection and classification. In the first stage, a number of discriminative features out of 18 were selected using PSO and several feature selection techniques to reduce the features dimension. In the second stage, we applied the random forest ensemble classification scheme to diagnose lymphatic diseases. While making experiments with the selected features, we used original and resampled distributions of the dataset to train random forest classifier. Experimental results demonstrate that the proposed method achieves a remark-able improvement in classification accuracy rate.

Ensemble Pruning for Glaucoma Detection in an Unbalanced Data Set.

Random forests are successful classifier ensemble methods consisting of typically 100 to 1000 classification trees. Ensemble pruning techniques reduce the computational cost, especially the memory demand, of random forests by reducing the number of trees without relevant loss of performance or even with increased performance of the sub-ensemble. The application to the problem of an early detection of glaucoma, a severe eye disease with low prevalence, based on topographical measurements of the eye background faces specific challenges. We examine the performance of ensemble pruning strategies for glaucoma detection in an unbalanced data situation. The data set consists of 102 topographical features of the eye background of 254 healthy controls and 55 glaucoma patients. We compare the area under the receiver operating characteristic curve (AUC), and the Brier score on the total data set, in the majority class, and in the minority class of pruned random forest ensembles obtained with strategies based on the prediction accuracy of greedily grown sub-ensembles, the uncertainty weighted accuracy, and the similarity between single trees. To validate the findings and to examine the influence of the prevalence of glaucoma in the data set, we additionally perform a simulation study with lower prevalences of glaucoma. In glaucoma classification all three pruning strategies lead to improved AUC and smaller Brier scores on the total data set with sub-ensembles as small as 30 to 80 trees compared to the classification results obtained with the full ensemble consisting of 1000 trees. In the simulation study, we were able to show that the prevalence of glaucoma is a critical factor and lower prevalence decreases the performance of our pruning strategies. The memory demand for glaucoma classification in an unbalanced data situation based on random forests could effectively be reduced by the application of pruning strategies without loss of performance in a population with increased risk of glaucoma.

Automated Identification of Abnormal Adult EEGs.

The interpretation of electroencephalograms (EEGs) is a process that is still dependent on the subjective analysis of the examiners. Though interrater agreement on critical events such as seizures is high, it is much lower on subtler events (e.g., when there are benign variants). The process used by an expert to interpret an EEG is quite subjective and hard to replicate by machine. The performance of machine learning technology is far from human performance. We have been developing an interpretation system, AutoEEG, with a goal of exceeding human performance on this task. In this work, we are focusing on one of the early decisions made in this process - whether an EEG is normal or abnormal. We explore two baseline classification algorithms: k-Nearest Neighbor (kNN) and Random Forest Ensemble Learning (RF). A subset of the TUH EEG Corpus was used to evaluate performance. Principal Components Analysis (PCA) was used to reduce the dimensionality of the data. kNN achieved a 41.8% detection error rate while RF achieved an error rate of 31.7%. These error rates are significantly lower than those obtained by random guessing based on priors (49.5%). The majority of the errors were related to misclassification of normal EEGs.

Abstract LB-177: Novel approaches for improving interpretation and predictive models of comparative genomic hybridization data

Abstract As costs of genome wide analyses decline and become more accessible, their use in both human and animal cancer studies are generating increasing information regarding underlying cancer etiology. Comparative genomic hybridization (CGH) is providing valuable information relating copy number aberrations (CNAs) to cancer mechanisms and clinical outcomes. However, challenges exist to interpreting and fully utilizing these data. First, without matched tumor and healthy tissue samples from individuals, distinguishing naturally occurring copy number variations (CNVs) from CNAs is difficult. Second, the large search space of genome wide analyses makes finding combinations of CNAs with improved predictive potential compared to single CNAs challenging. Here we provide novel methods to address these challenges associated with CGH data. Many new resources (e.g. The Cancer Genome Atlas (TCGA)), are making large volumes of genomic data publically accessible. However, most datasets do not have matched normal and tumor tissue samples between subjects. We tested matched normal and tissue samples from 30 patients with colorectal, lung, and pancreatic cancer and compared CNVs and CNAs to findings in larger, non-matched samples in TCGA. Even with limited matched samples, this approach allows for the differentiation of CNVs from CNAs discovered in analyses of non-matched samples. In some cases, combinations of CNAs can provide improved predictive capability compared to any single CNA. However, it is computationally intractable to exhaustively test combinations of CNAs in a genome-wide study. To address this limitation, we use a novel approach for CNA feature reduction that minimizes the variance within CNA segments across subjects, and Random Forest Ensemble Classification. This approach provides CNA combinations with balanced accuracies of 83.5% and 94.9% for distinguishing 52 cases of canine ALL/AML and 71 cases of B-CLL/T-CLL, respectively. These two approaches address frequent limitations in the interpretation CGH data. Better distinguishing CNVs and interrogating CNA combinations, can provide additional information about the role of CNAs in disease mechanisms and improve treatment decisions. Citation Format: Daniel Rotroff, Matthew Breen, Alison Motsinger-Reif. Novel approaches for improving interpretation and predictive models of comparative genomic hybridization data. [abstract]. In: Proceedings of the 106th Annual Meeting of the American Association for Cancer Research; 2015 Apr 18-22; Philadelphia, PA. Philadelphia (PA): AACR; Cancer Res 2015;75(15 Suppl):Abstract nr LB-177. doi:10.1158/1538-7445.AM2015-LB-177

Predicting Eucalyptus spp. stand volume in Zululand, South Africa: an analysis using a stochastic gradient boosting regression ensemble with multi-source data sets

Accurate, reliable, and up-to-date forest stand volume information is a prerequisite for a detailed evaluation of commercial forest resources and their sustainable management. Commercial forest responses to global climate change remain uncertain, and hence the mapping of stand volume as carbon sinks is fundamentally important in understanding the role of forests in stabilizing climate change effects. The aim of this study was to examine the utility of stochastic gradient boosting (SGB) and multi-source data to predict stand volume of a Eucalyptus plantation in South Africa. The SGB ensemble, random forest (RF), and stepwise multiple-linear regression (SMLR) were used to predict Eucalyptus stand volume and other related tree-structural attributes such as mean tree height and mean diameter at breast height (DBH). Multi-source data consisted of SPOT-5 raw spectral features (four bands), 14 spectral vegetation indices, rainfall data, and stand age. When all variables were used, the SGB algorithm showed that stand volume can be accurately estimated (R2 = 0.78 and RMSE = 33.16 m3 ha−1 (23.01% of the mean)). The competing RF ensemble produced an R2 value of 0.76 and a RMSE value of 37.28 m3 ha−1 (38.28% of the mean). SMLR on the other hand, produced an R2 value of 0.65 and an RMSE value of 42.50 m3 ha−1 (42.50% of the mean). Our study further showed that Eucalyptus mean tree height (R2 = 0.83 and RMSE = 1.63 m (9.08% of the mean)) and mean diameter at breast height (R2 = 0.74 and RMSE = 1.06 (7.89% of the mean)) can also be reasonably predicted using SGB and multi-source data. Furthermore, when the most important SGB model-selected variables were used for prediction, the predictive accuracies improved significantly for mean DBH (R2 = 0.81 and RMSE = 1.21 cm (6.12% of the mean)), mean tree height (R2 = 0.86 and RMSE = 1.39 m (7.02% of the mean)), and stand volume (R2 = 0.83 and RMSE = 29.58 m3 ha−1 (17.63% of the mean)). These results underscore the importance of integrating multi-source data with remotely sensed data for predicting Eucalyptus stand volume and related tree-structural attributes.

Object-based detection of Arctic sea ice and melt ponds using high spatial resolution aerial photographs

High resolution aerial photographs used to detect and classify sea ice features can provide accurate physical parameters to refine, validate, and improve climate models. However, manually delineating sea ice and melt ponds is time-consuming and labor-intensive. In this study, an object-based classification algorithm is developed to automatically extract sea ice and melt ponds efficiently from 163 aerial photographs taken during the Chinese National Arctic Research Expedition in summer 2010 (CHINARE 2010) in the Arctic Pacific Sector. The photographs are selected from 599 cloud-free photographs based on their image quality and representativeness in the marginal ice zone (MIZ). The algorithm includes three major steps: (1) the image segmentation groups the neighboring pixels into objects according to the similarity of spectral and textural information; (2) the random forest ensemble classifier distinguishes four general classes: water, general submerged ice (GSI, including melt ponds and submerged ice along ice edges), shadow, and ice/snow; and (3) the polygon neighbor analysis further separates melt ponds and submerged ice from the GSI according to their spatial relationships. The overall classification accuracy for the four general classes is 95.5% based on 178 ground reference objects. Furthermore, the producer's accuracy of 90.8% and user's accuracy of 91.8% are achieved for melt pond detection through 98 independent reference objects. For the 163 photos examined, a total of 19,438 melt ponds larger than 1m2 are detected, with a pond density of 867.2km−2, mean pond size of 32.6±0.03m2, and mean pond fraction of 0.06±0.006; a total of 42,468 ice floes are detected, with the mean floe size of 173.3±0.1m2 (majority in 1–30m2) and mean ice concentration of 46.1±0.5% (ranging from 18.6–98.6%). These results matched well with ship-based visual observations in the MIZ in the same area and time. The method presented in the paper can be applied to data sets of high spatial resolution Arctic sea ice photographs for deriving detailed sea ice concentration, floe size, and melt pond distributions over wider regions, and extracting sea ice physical parameters and their corresponding changes between years.

Pedestrian detection based on hierarchical co-occurrence model for occlusion handling

In pedestrian detection, occlusions are typically treated as an unstructured source of noise and explicit models have lagged behind those for object appearance, which will result in degradation of detection performance. In this paper, a hierarchical co-occurrence model is proposed to enhance the semantic representation of a pedestrian. In our proposed hierarchical model, a latent SVM structure is employed to model the spatial co-occurrence relations among the parent–child pairs of nodes as hidden variables for handling the partial occlusions. Moreover, the visibility statuses of the pedestrian can be generated by learning co-occurrence relations from the positive training data with large numbers of synthetically occluded instances. Finally, based on the proposed hierarchical co-occurrence model, a pedestrian detection algorithm is implemented to incorporate visibility statuses by means of a Random Forest ensemble. The experimental results on three public datasets demonstrate the log-average miss rate of the proposed algorithm has 5% improvement for pedestrians with partial occlusions compared with the state-of-the-arts.

Non-intrusive appliance load monitoring with bagging classifiers

Nonintrusive appliance load monitoring is an important problem class with interesting applications.Due to the diculties that arise from the application of data mining techniques to real-world datasets, we close a gap in literature and focus on robust bagging classi ers. In this work, we answer thequestion, if the recognition rate of ensemble classi ers is signi cantly better than the recognition rateof the native classi ers. We analyze two types of bagging classi ers, i.e., (1) support vector machineand nearest neighbor ensembles and (2) random forests. We compare their performance in terms ofaccuracy and robustness on a NIALM data set recorded in a eld study. The experimental analysisconcentrates on recognition rates w.r.t. various training set sizes, on the inuence of neighborhoodsizes and the numbers of decision trees in random forest ensembles. It turns out that the decisiontree ensembles belong to the best classi ers in the employed scenarios.Keywords: ensembles, NIALM, support vector machines, nearest neighbor classi cation, randomforests

Learning from automatically labeled data: case study on click fraud prediction

In the era of big data, both class labels and covariates may result from proprietary algorithms or ground models. The predictions of these ground models, however, are not the same as the unknown ground truth. Thus, the automatically generated class labels are inherently uncertain, making subsequent supervised learning from such data a challenging task. Fine-tuning a new classifier could mean that, at the extreme, this new classifier will try to replicate the decision heuristics of the ground model. However, few new insights can be expected from a model that tries to merely emulate another one. Here, we study this problem in the context of click fraud prediction from highly skewed data that were automatically labeled by a proprietary detection algorithm. We propose a new approach to generate click profiles for publishers of online advertisements. In a blinded test, our ensemble of random forests achieved an average precision of only 36.2 %, meaning that our predictions do not agree very well with those of the ground model. We tried to elucidate this discrepancy and made several interesting observations. Our results suggest that supervised learning from automatically labeled data should be complemented by an interpretation of conflicting predictions between the new classifier and the ground model. If the ground truth is not known, then elucidating such disagreements might be more relevant than improving the performance of the new classifier.

Ara h 1 Peptide Immunotherapy Ameliorates Peanut-Induced Anaphylaxis

S U N D A Y 515 Identifying CpG Sites Associated with Eczema Via Random Forest Screening of Epigenome-Wide DNA Methylation Bilal M. Quraishi, Hongmei Zhang, PhD, Todd M. Everson, Gabrielle A. Lockett, PhD, Meredith Ray, John W. Holloway, PhD, Syed H. Arshad, DM, FRCP, Wilfried Karmaus, MD, DrMed, MPH; University of Memphis, Memphis, TN, University of South Carolina, Columbia, SC, University of Southampton, Southampton, United Kingdom, University of Southampton, United Kingdom, The David Hide Asthma and Allergy Research Centre, United Kingdom. RATIONALE: The prevalence of eczema among children of industrialized countries is increasing. There is some evidence that DNAmethylation plays a role in eczema, however, there is lack of epi-genome-wide study on their association. METHODS: Association between eczema and DNA methylation was studied in 18year-old subjects (n5366) from the Isle of Wight (IOW) birth Cohort. To efficiently screen cytosine-phospate-guanine (CpG) sites in epi-genome-wide, we applied the approach of random forest (RF) ensemble. A total of 307,357 CpGs were subjected to random forest data reduction, repeatedly dropping 50% of variables with lowest variable importance measures until the misclassification rate showed a significant increase. Functional annotation and pathway analyses along with logistic regressions were performed to further evaluate the selected RF-CpG sites. RESULTS: The RF method yielded 75 CpGs, 72 of which are linked to eczema corroborated by logistic regression. Eczema-associated CpGs, e.g., in genes PARD3 and GUCY1A3, were significantly enriched within the pathways ‘Tight junction’ and ‘Gap junction’ (P values 0.0008 and 0.002 respectively, after correcting for multiple testing with false discovery rate of 0.05). It has been shown that genes in these pathways are correlated to allergic and immune related diseases. CONCLUSIONS: TheRFEnsemblemethodwas successfully utilized for epigenome-wide scanning to identify epigenetic loci associated with eczema, detecting both previously known and novel CpG sites. The significant enrichment of differentially methylated CpGs within tight and gap junction pathway genes implicates epigenetic regulation in epidermal barrier dysfunction in eczema.

A tool for identifying potential Eucalyptus nitens seed orchard sites based on climate and topography

Shy seed production in orchards of Eucalyptus nitens is a major barrier to the deployment of genetic gain in South African plantations. A machine learning method was used to identify optimal sites for the establishment of E. nitens seed orchards within the plantation forestry landscape of the summer rainfall region of South Africa. The ensemble classifier random forests (RF) was used to identify the environmental factors conducive to E. nitens floral bud production, and, based on these, build a predictive model deployable to the plantation forestry landscape for identifying suitable areas for E. nitens seed orchards. The RF model predicted site suitability likelihood for floral bud production with a high level of accuracy (area under the receiver operating characteristic curve = 0.83). Within the climatically optimal range for growing E. nitens, flower bud production was more abundant and consistent on cold slopes, i.e. sites experiencing lower minimum air temperatures during spring and autumn. The model was applied to the commercial plantation forestry landscape for the purpose of indicating sites climatically optimal for floral bud production in E. nitens and the establishment of breeding and seed production orchards of the same species.

Evaluating the utility of the medium-spatial resolution Landsat 8 multispectral sensor in quantifying aboveground biomass in uMgeni catchment, South Africa

Aboveground biomass estimation is critical in understanding forest contribution to regional carbon cycles. Despite the successful application of high spatial and spectral resolution sensors in aboveground biomass (AGB) estimation, there are challenges related to high acquisition costs, small area coverage, multicollinearity and limited availability. These challenges hamper the successful regional scale AGB quantification. The aim of this study was to assess the utility of the newly-launched medium-resolution multispectral Landsat 8 Operational Land Imager (OLI) dataset with a large swath width, in quantifying AGB in a forest plantation. We applied different sets of spectral analysis (test I: spectral bands; test II: spectral vegetation indices and test III: spectral bands+spectral vegetation indices) in testing the utility of Landsat 8 OLI using two non-parametric algorithms: stochastic gradient boosting and the random forest ensembles. The results of the study show that the medium-resolution multispectral Landsat 8 OLI dataset provides better AGB estimates for Eucalyptus dunii, Eucalyptus grandis and Pinus taeda especially when using the extracted spectral information together with the derived spectral vegetation indices. We also noted that incorporating the optimal subset of the most important selected medium-resolution multispectral Landsat 8 OLI bands improved AGB accuracies. We compared medium-resolution multispectral Landsat 8 OLI AGB estimates with Landsat 7 ETM+estimates and the latter yielded lower estimation accuracies. Overall, this study demonstrates the invaluable potential and strength of applying the relatively affordable and readily available newly-launched medium-resolution Landsat 8 OLI dataset, with a large swath width (185-km) in precisely estimating AGB. This strength of the Landsat OLI dataset is crucial especially in sub-Saharan Africa where high-resolution remote sensing data availability remains a challenge.

Performance-weighted ensembles of random forests for predicting price impact

For any large player in financial markets, the impact of their trading activity represents a substantial proportion of transaction costs. This paper proposes a novel machine learning algorithm for predicting the price impact of order book events. Specifically, we introduce a prediction system based on ensembles of random forests (RFs). The system is trained and tested on depth-of-book data from the BATS and Chi-X exchanges and performance is benchmarked using ensembles of other popular regression algorithms including: linear regression, neural networks and support vector regression. The results show that recency-weighted ensembles of RFs produce over 15% greater prediction accuracy on out-of-sample data, for 5 out of 6 timeframes studied, compared with all benchmarks. Feature importance ranking is used to explore the significance of various market features on the price impact, finding them to be highly variable through time. Finally, a novel procedure for extracting the directional effects of features is proposed and used to explore the features most dominant in the price formation process.

데이터 전처리와 앙상블 기법을 통한 불균형 데이터의 분류모형 비교 연구

There are many studies related to imbalanced data in which the class distribution is highly skewed. To address the problem of imbalanced data, previous studies deal with resampling techniques which correct the skewness of the class distribution in each sampled subset by using under-sampling, over-sampling or hybridsampling such as SMOTE. Ensemble methods have also alleviated the problem of class imbalanced data. In this paper, we compare around a dozen algorithms that combine the ensemble methods and resampling techniques based on simulated data sets generated by the Backbone model, which can handle the imbalance rate. The results on various real imbalanced data sets are also presented to compare the e.ectiveness of algorithms. As a result, we highly recommend the resampling technique combining ensemble methods for imbalanced data in which the proportion of the minority class is less than 10%. We also .nd that each ensemble method has a well-matched sampling technique. The algorithms which combine bagging or random forest ensembles with random undersampling tend to perform well; however, the boosting ensemble appears to perform better with over-sampling. All ensemble methods combined with SMOTE outperform in most situations.

Identification of non-coding RNAs with a new composite feature in the Hybrid Random Forest Ensemble algorithm.

To identify non-coding RNA (ncRNA) signals within genomic regions, a classification tool was developed based on a hybrid random forest (RF) with a logistic regression model to efficiently discriminate short ncRNA sequences as well as long complex ncRNA sequences. This RF-based classifier was trained on a well-balanced dataset with a discriminative set of features and achieved an accuracy, sensitivity and specificity of 92.11%, 90.7% and 93.5%, respectively. The selected feature set includes a new proposed feature, SCORE. This feature is generated based on a logistic regression function that combines five significant features—structure, sequence, modularity, structural robustness and coding potential—to enable improved characterization of long ncRNA (lncRNA) elements. The use of SCORE improved the performance of the RF-based classifier in the identification of Rfam lncRNA families. A genome-wide ncRNA classification framework was applied to a wide variety of organisms, with an emphasis on those of economic, social, public health, environmental and agricultural significance, such as various bacteria genomes, the Arthrospira (Spirulina) genome, and rice and human genomic regions. Our framework was able to identify known ncRNAs with sensitivities of greater than 90% and 77.7% for prokaryotic and eukaryotic sequences, respectively. Our classifier is available at http://ncrna-pred.com/HLRF.htm.

Condition monitoring of induction motors: A review and an application of an ensemble of hybrid intelligent models

In this paper, a review on condition monitoring of induction motors is first presented. Then, an ensemble of hybrid intelligent models that is useful for condition monitoring of induction motors is proposed. The review covers two parts, i.e., (i) a total of nine commonly used condition monitoring methods of induction motors; and (ii) intelligent learning models for condition monitoring of induction motors subject to single and multiple input signals. Based on the review findings, the Motor Current Signature Analysis (MCSA) method is selected for this study owing to its online, non-invasive properties and its requirement of only single input source; therefore leading to a cost-effective condition monitoring method. A hybrid intelligent model that consists of the Fuzzy Min–Max (FMM) neural network and the Random Forest (RF) model comprising an ensemble of Classification and Regression Trees is developed. The majority voting scheme is used to combine the predictions produced by the resulting FMM–RF ensemble (or FMM–RFE) members. A benchmark problem is first deployed to evaluate the usefulness of the FMM–RFE model. Then, the model is applied to condition monitoring of induction motors using a set of real data samples. Specifically, the stator current signals of induction motors are obtained using the MCSA method. The signals are processed to produce a set of harmonic-based features for classification using the FMM–RFE model. The experimental results show good performances in both noise-free and noisy environments. More importantly, a set of explanatory rules in the form of a decision tree can be extracted from the FMM–RFE model to justify its predictions. The outcomes ascertain the effectiveness of the proposed FMM–RFE model in undertaking condition monitoring tasks, especially for induction motors, under different environments.

High-resolution landcover classification using Random Forest

Potential data sets for landcover classification, such as Landsat (or pre-processed data such as the National Land Cover Dataset (NLCD)), are often too coarse for fine-scale research needs or are cost-prohibitive (Quickbird, Ikonos and Geoeye). Repeated attempts at classifying high spatial resolution data, National Agricultural Imagery Program (NAIP) imagery, based on traditional techniques, such as a maximum likelihood supervised classification, have failed to produce a product with sufficient accuracy. We used the ensemble Random Forests (RFs) classifier to classify landcover at 1 m resolution using 2009 NAIP imagery in south-eastern Wyoming. We classified riparian areas within a 225 km2 study area, at 1 m spatial resolution, using RFs with emphasis on riparian corridors that yielded a land cover map with overall accuracy of 81% and a kappa coefficient of 79%. Users’ accuracy of important riparian vegetation species, aspen, riparian grasses and willow were 79%, 81% and 83%, respectively. Techniques presented in this paper, which exploit free NAIP imagery for landcover classification, represent an inexpensive and reliable alternative to purchasing commercial imagery when high spatial resolution landcover data is needed.

Predicting motor learning performance from Electroencephalographic data

BackgroundResearch on the neurophysiological correlates of visuomotor integration and learning (VMIL) has largely focused on identifying learning-induced activity changes in cortical areas during motor execution. While such studies have generated valuable insights into the neural basis of VMIL, little is known about the processes that represent the current state of VMIL independently of motor execution. Here, we present empirical evidence that a subject’s performance in a 3D reaching task can be predicted on a trial-to-trial basis from pre-trial electroencephalographic (EEG) data. This evidence provides novel insights into the brain states that support successful VMIL.MethodsSix healthy subjects, attached to a seven degrees-of-freedom (DoF) robot with their right arm, practiced 3D reaching movements in a virtual space, while an EEG recorded their brain’s electromagnetic field. A random forest ensemble classifier was used to predict the next trial’s performance, as measured by the time needed to reach the goal, from pre-trial data using a leave-one-subject-out cross-validation procedure.ResultsThe learned models successfully generalized to novel subjects. An analysis of the brain regions, on which the models based their predictions, revealed areas matching prevalent motor learning models. In these brain areas, the α/μ frequency band (8–14 Hz) was found to be most relevant for performance prediction.ConclusionsVMIL induces changes in cortical processes that extend beyond motor execution, indicating a more complex role of these processes than previously assumed. Our results further suggest that the capability of subjects to modulate their α/μ bandpower in brain regions associated with motor learning may be related to performance in VMIL. Accordingly, training subjects in α/μ-modulation, e.g., by means of a brain-computer interface (BCI), may have a beneficial impact on VMIL.

Automated trading with performance weighted random forests and seasonality

Seasonality effects and empirical regularities in financial data have been well documented in the financial economics literature for over seven decades. This paper proposes an expert system that uses novel machine learning techniques to predict the price return over these seasonal events, and then uses these predictions to develop a profitable trading strategy. While simple approaches to trading these regularities can prove profitable, such trading leads to potential large drawdowns (peak-to-trough decline of an investment measured as a percentage between the peak and the trough) in profit. In this paper, we introduce an automated trading system based on performance weighted ensembles of random forests that improves the profitability and stability of trading seasonality events. An analysis of various regression techniques is performed as well as an exploration of the merits of various techniques for expert weighting. The performance of the models is analysed using a large sample of stocks from the DAX. The results show that recency-weighted ensembles of random forests produce superior results in terms of both profitability and prediction accuracy compared with other ensemble techniques. It is also found that using seasonality effects produces superior results than not having them modelled explicitly.