Improve Prediction Results Research Articles

Augmented deep learning workflow for robust fault prediction over multiple tectonic regimes

Abstract Over the past decade, deep learning frameworks such as convolutional neural networks (CNNs) have made major research inroads in upstream oil and gas, with applications in seismic processing/imaging, velocity model building, petrophysics, geological seismic interpretation, all the way to development, production and supply chain logistics. CNN fault prediction centers around the idea of image edge detection and, for improved prediction results, three data-driven steps are recommended. First, pre-condition the seismic data to increase signal-to-noise ratio as much as possible: iterative dip-steered median filtering and principal component filtering are adopted to further improve signal-to-noise ratio and sharpness for edge detection. Second, a fault probability volume obtained through deep learning (DL) -based fault detection using a U-Net architecture, taking synthetic seismic models as samples that can improve DL-based fault prediction in lieu of readily available labeled fault sets that may be prohibitively time-consuming to generate. Finally, edge enhancement is performed on the inference results to improve precision and fault continuity. A comparative analysis between related edge enhancement technologies is also presented. Results for three different faulting modes (normal, reverse and strike-slip) over three real seismic field data sets from China demonstrate the robustness of the proposed workflow.

Decrypting the information captured by MRI-radiomic features in predicting the response to neoadjuvant chemotherapy in breast cancer.

MRI-based radiomic models have shown promises in predicting the response to neoadjuvant chemotherapy in breast cancer. However, it is difficult to determine which information from the images contributes the most to the prediction: the distribution of gray-levels, the tumour heterogeneity, the shape of the lesions or the intensities of peritumoural regions. The purpose of this study is to dissociate the different sources of information to improve prediction results. Based on pre-treatment MR images from 103 patients, four types of 3D Volumes Of Interest were defined and arranged in multiple combinations. Combining features extracted from different regions proved to increase prediction performances. Clinical relevance- This study proposes a method based on analyses of MRI tumor heterogeneity, margins and peritumoral regions to improve the prediction of the response to neoadjuvant chemotherapy in breast cancer, which would help personalize patient treatment.

Predicting Soil Properties and Interpreting Vis-NIR Models from across Continental United States.

The United States NRCS has a soil database that has data collected from across the country over the last several decades. This also includes soil spectral scans. This data is available, but it may not be used to its full potential. For this study, pedon, horizon and spectral data was extracted from the database for samples collected from 2011 to 2015. Only sites that had been fully described and horizons that had been analyzed for the full suite of desired properties were used. This resulted in over 14,000 samples that were used for modeling and eight soil properties: soil organic carbon (SOC); total nitrogen (TN); total sulfur (TS); clay; sand; exchangeable calcium (Caex); cation exchange capacity (CEC); and pH. Four chemometric methods were employed for soil property prediction: partial least squares (PLSR); Random Forest (RF); Cubist; and multivariable adaptive regression splines (MARS). The dataset was sufficiently large that only random subsetting was used to create calibration (70%) and validation (30%) sets. SOC, TN, and TS had the strongest prediction results, with an R2 value of over 0.9. Caex, CEC and pH were predicted moderately well. Clay and sand models had slightly lower performance. Of the four methods, Cubist produced the strongest models, while PLSR produced the weakest. This may be due to the complex relationships between soil properties and spectra that PLSR could not capture. The only drawback of Cubist is the difficult interpretability of variable importance. Future research should include the use of environmental variables to improve prediction results. Future work may also avoid the use of PLSR when dealing with large datasets that cover large areas and have high degrees of variability.

Traffic speed prediction of high‐frequency time series using additively decomposed components as features

AbstractTraffic speed prediction is an integral part of an Intelligent Transportation System (ITS) and the Internet of Vehicles (IoV). Advanced knowledge of average traffic speed can help take proactive preventive steps to avoid impending problems. There have been studies for traffic speed prediction in which data has been decomposed into components using various decomposition techniques such as empirical mode decomposition, wavelets, and seasonal decomposition. As far as the authors are aware, no research has used additively decomposed components as input features. In this study, we used additive decomposition on 21,843 samples of traffic speed data. We implemented two statistical techniques designed for double seasonality (i) Double Seasonal Holt‐Winter, and (ii) Trigonometric seasonality, Box‐Cox transformation, autoregressive integrated moving average errors, trend, and Seasonal components (TBATS), and five machine learning (ML) techniques, (i) Multi‐Layer Perceptron, (ii) Convolutional‐Neural Network, (iii) Long Short‐Term Memory, (iv) Gated Recurrent Unit and (v) Convolutional‐Neural Network‐LSTM. Machine learning techniques are used in univariate mode with raw time series as features and then with decomposed components as features in multivariate mode. This study demonstrates that using decomposed components (trend, seasonal, and residual), as features, improves prediction results for multivariate ML techniques. This becomes a significant advantage when no other features are available.

Application of Neural Networks to Predict Daecheong Dam Water Levels

The accurate prediction of water levels is essential for the efficient operation of hydraulic structures such as dams. However, such predictions are impossible when considering all factors affecting water level changes. Recently, with the development of artificial neural networks, there has been research on water level prediction using deep learning models. In this study, the deep learning technique of Multi-Layer Perceptron (MLP) was combined with Recurrent Neural Network (RNN) and Long Short-Term Memory (LSTM) techniques, which afford advantages in time series data processing to predict the water level of Daecheong Dam. Three data preprocessing processes were applied to improve accuracy in water level predictions by taking travel time into account, adding the outflow data of the dam, and normalizing the data. Eight cases were created with the three data preprocessing processes. When normalization was performed for the data that took travel time into account and the data were then used as input data for LSTM, the best prediction result was found for the peak water level of the dam. If various other data preprocessing techniques are applied in addition to those applied in this study, RNN and LSTM will show improved prediction results.

Uncertainty analysis and calibration of SST turbulence model for free shear layer in cavity-ramp flow

The development and reattachment of the free shear layer in supersonic flows have been a widely encountered phenomenon in engineering, and are gaining increasing interest from researchers who investigate such complex flow structures using numerical simulation methods. Among these methods, Reynolds Averaged Navier-Stokes (RANS) still plays an important role in application problems. However, the uncertainty of the parameters in RANS turbulence model affects the reliability and accuracy of predictions. Therefore, the purpose of this study is to conduct a detailed uncertainty analysis of the parameters in the shear-stress transport (SST) turbulence model for the free shear layer in a cavity-ramp flow. In the current study, uncertainty quantification was conducted with the construction of a surrogate model between the parameters and calculation results using the non-intrusive polynomial chaos (NIPC) method. Next, for sensitivity analysis, the Sobol index of each parameter was calculated to identify the key parameters mainly responsible for the uncertainty. Based on the above results, Bayesian inference was employed to calibrate these key parameters by leveraging experimental data. The results show that the calculated wall pressure and skin friction of the ramp suffer a non-negligible uncertainty caused by the model parameters, and parameters a1, σω2, β2, κ, and β1 are identified as the key parameters in this type of flow. Subsequently, the calibrated values for the key parameters are obtained through Bayesian inference, which could significantly improve prediction results in the original case. Furthermore, the applicability of the SST model with calibrated parameters is verified under different free-stream conditions.

Geoinformation-based landslide susceptibility mapping in subtropical area

Mapping susceptibility of landslide disaster is essential in subtropical area, where abundant rainfall may trigger landslide and mudflow, causing damages to human society. The purpose of this paper is to propose an integrated methodology to achieve such a mapping work with improved prediction results using hybrid modeling taking Chongren, Jiangxi as an example. The methodology is composed of the optimal discretization of the continuous geo-environmental factors based on entropy, weight of evidence (WoE) calculation and application of the known machine learning (ML) models, e.g., Random Forest (RF), Support Vector Machine (SVM) and Logistic Regression (LR). The results show the effectiveness of the proposed hybrid modeling for landslide hazard mapping in which the prediction accuracy vs the validation set reach 82.35–91.02% with an AUC [area under the receiver operating characteristic (ROC) curve] of 0.912–0.970. The RF algorithm performs best among the observed three ML algorithms and WoE-based RF modeling will be recommended for the similar landslide risk prediction elsewhere. We believe that our research can provide an operational reference for predicting the landslide hazard in the subtropical area and serve for disaster reduction and prevention action of the local governments.

PSO Based Optimized Ensemble Learning and Feature Selection Approach for Efficient Energy Forecast

Swarm intelligence techniques with incredible success rates are broadly used for various irregular and interdisciplinary topics. However, their impact on ensemble models is considerably unexplored. This study proposes an optimized-ensemble model integrated for smart home energy consumption management based on ensemble learning and particle swarm optimization (PSO). The proposed model exploits PSO in two distinct ways; first, PSO-based feature selection is performed to select the essential features from the raw dataset. Secondly, with larger datasets and comprehensive range problems, it can become a cumbersome task to tune hyper-parameters in a trial-and-error manner manually. Therefore, PSO was used as an optimization technique to fine-tune hyper-parameters of the selected ensemble model. A hybrid ensemble model is built by using combinations of five different baseline models. Hyper-parameters of each combination model were optimized using PSO followed by training on different random samples. We compared our proposed model with our previously proposed ANN-PSO model and a few other state-of-the-art models. The results show that optimized-ensemble learning models outperform individual models and the ANN-PSO model by minimizing RMSE to 6.05 from 9.63 and increasing the prediction accuracy by 95.6%. Moreover, our results show that random sampling can help improve prediction results compared to the ANN-PSO model from 92.3% to around 96%.

Ensemble of Multiple Classifiers for Multilabel Classification of Plant Protein Subcellular Localization.

The accurate prediction of protein localization is a critical step in any functional genome annotation process. This paper proposes an improved strategy for protein subcellular localization prediction in plants based on multiple classifiers, to improve prediction results in terms of both accuracy and reliability. The prediction of plant protein subcellular localization is challenging because the underlying problem is not only a multiclass, but also a multilabel problem. Generally, plant proteins can be found in 10–14 locations/compartments. The number of proteins in some compartments (nucleus, cytoplasm, and mitochondria) is generally much greater than that in other compartments (vacuole, peroxisome, Golgi, and cell wall). Therefore, the problem of imbalanced data usually arises. Therefore, we propose an ensemble machine learning method based on average voting among heterogeneous classifiers. We first extracted various types of features suitable for each type of protein localization to form a total of 479 feature spaces. Then, feature selection methods were used to reduce the dimensions of the features into smaller informative feature subsets. This reduced feature subset was then used to train/build three different individual models. In the process of combining the three distinct classifier models, we used an average voting approach to combine the results of these three different classifiers that we constructed to return the final probability prediction. The method could predict subcellular localizations in both single- and multilabel locations, based on the voting probability. Experimental results indicated that the proposed ensemble method could achieve correct classification with an overall accuracy of 84.58% for 11 compartments, on the basis of the testing dataset.

Abstract IA-23: Multi-scale modeling of cancer patients

Abstract Recently biomedical analysis in both imaging and non-imaging data has advanced rapidly aided by the modeling power of deep learning networks. Complex diseases such as cancer may reveal different aspects of disease characteristics in different data modalities, and thus integrating imaging with non-imaging biomarkers and clinical data has the potential to discover information that is not obvious in any single data modality. We are developing methods for integrating image, genomic and clinical data to improve precision medicine applications for cancer patients. Our results suggest that the integration strategies achieve improved prediction results than using each single data modality alone. We further show that information is transferrable across age groups or tumor subtypes under the data fusion framework. Citation Format: Olivier Gevaert. Multi-scale modeling of cancer patients [abstract]. In: Proceedings of the AACR Virtual Special Conference on Artificial Intelligence, Diagnosis, and Imaging; 2021 Jan 13-14. Philadelphia (PA): AACR; Clin Cancer Res 2021;27(5_Suppl):Abstract nr IA-23.

Achieving robustness across season, location and cultivar for a NIRS model for intact mango fruit dry matter content

Abstract Short wave near infrared spectroscopy has found use in non-invasive assessment of dry matter content (DMC, % fresh weight) of mango fruit, both as a guide to harvest maturity and ensure eating quality of ripened fruit. In this study, this application is optimised in terms of pre-processing of spectra, the source of variations important to model performance documented, and the performance of cultivar or physiological stage specific partial least squares regression (PLSR) models, global PLSR and an artificial neural network (ANN) model are compared. The data set consisted of 4675 samples acquired across four seasons, ten cultivars and two growing regions, with harvest populations used as cross validation groups. The data of the fourth season was reserved as an independent test set. Spectra pre-treatment of mean centred Savitzy-Golay second derivative (second order polynomial using a 17 point interval) and use of the wavelength range 684−990 nm gave the lowest RMSECV for PLSR models, although other ranges had similar statistics. The fruit physiological stage had the greatest impact on PLSR model performance, compared to cultivar, year or growing region, as estimated using a ‘variable importance metric’ devised and implemented using a random forest regression. The use of specific (cultivar or physiological stage) PLSR models improved prediction results of the independent validation set (RMSEP on DMC decreased from 1.01 to 0.88 %), and was similar to the result of a global ANN model (0.89 %). The use of an ANN model is recommended in terms of ease of use of a single model across all cultivars.

An enhanced computer-assisted lung cancer detection method using content based image retrieval and data mining techniques

Cancer is a sickness brought about by an uncontrolled division of eccentric cells in any part of human body. It is in the top of few places in the killer disease list and pervades in the entire world, but still on the rise. Most of the cases an early detection of lung cancer is cumbersome. This research paper is aimed to present an effective and an efficient way of computer-assisted detection method for lung cancer. In this research we used a set of lung computed tomography scanned images as inputs, obtained from lung image archives and applied image processing techniques such as feature extraction, segmentation. In this approach, a proper combination of Adaptive thresholding segmentation algorithm has been used for segmenting input images, a well-known Support Vector Machine image classification algorithm has been used for lung tumor classification and Content-based image retrieval technique has been used to compare lung image features such as contract, intensity, texture and shape. A set of patient personal data is included to get more accurate and correct prediction results, and it is dealt with data mining approach. The proposed segmentation method shows improved prediction results.

Label-free diagnosis for colorectal cancer through coffee ring-assisted surface-enhanced Raman spectroscopy on blood serum.

Surface-enhanced Raman spectroscopy (SERS) is garnering considerable attention for the swift diagnosis of pathogens and abnormal biological status, that is, cancers. In this work, a simple, fast and inexpensive optical sensing platform is developed by the design of SERS sampling and data analysis. The pretreatment of spectral measurement employed gold nanoparticle colloid mixing with the serum from patients with colorectal cancer (CRC). The droplet of particle-serum mixture formed coffee-ring-like region at the rim, providing strong and stable SERS profiles. The obtained spectra from cancer patients and healthy volunteers were analyzed by unsupervised principal component analysis (PCA) and supervised machine learning model, such as support-vector machine (SVM), respectively. The results demonstrate that the SVM model provides the superior performance in the classification of CRC diagnosis compared with PCA. In addition, the values of carcinoembryonic antigen from the blood samples were compiled with the corresponding SERS spectra for SVM calculation, yielding improved prediction results.

SNOMED CT-Based Standardized e-Clinical Pathways for Enabling Big Data Analytics in Healthcare

Automation of healthcare facilities represents a challenging task of streamlining a highly information-intensive sector. Modern healthcare processes produce large amounts of data that have great potential for health policymakers and data science researchers. However, a considerable portion of such data is not captured in electronic format and hidden inside the paperwork. A major source of missing data in healthcare is paper-based clinical pathways (CPs). CPs are healthcare plans that detail the interventions for the treatment of patients, and thus are the primary source for healthcare data. However, most CPs are used as paper-based documents and not fully automated. A key contribution towards the full automation of CPs is their proper computer modeling and encoding their data with international clinical terminologies. We present in this research an ontology-based CP automation model in which CP data are standardized with SNOMED CT, thus enabling machine learning algorithms to be applied to CP-based datasets. CPs automated under this model contribute significantly to reducing data missingness problems, enabling detailed statistical analyses on CP data, and improving the results of data analytics algorithms. Our experimental results on predicting the Length of Stay (LOS) of stroke patients using a dataset resulting from an e-clinical pathway demonstrate improved prediction results compared with LOS prediction using traditional EHR-based datasets. Fully automated CPs enrich medical datasets with more CP data and open new opportunities for machine learning algorithms to show their full potential in improving healthcare, reducing costs, and increasing patient satisfaction.

Improved Deep Belief Network for Short-Term Load Forecasting Considering Demand-Side Management

Demand-side management (DSM) increases the complexity of forecasting environment, which makes traditional forecasting methods difficult to meet the firm's need for predictive accuracy. Since deep learning can comprehensively consider various factors to improve prediction results, this paper improves the deep belief network from three aspects of input data, model and performance, and uses it to solve the short-term load forecasting problem in DSM. In the data optimization stage, the Hankel matrix is constructed to increase the input weight of DSM data, and the gray relational analysis is used to select strongly correlated data from the data set. In the model optimization stage, the Gauss-Bernoulli restricted Boltzmann machine is used as the first restricted Boltzmann machine of the deep network to convert the continuity feature of input data into binomial distribution feature. In the performance optimization stage, a pre-training method combining error constraint and unsupervised learning is proposed to provide good initial parameters, and the global fine-tuning of network parameters is realized based on the genetic algorithm. Based on the actual data of Tianjin Power Grid in China, the experimental results show that the proposed method is superior to other methods.

Rough set-based feature selection for credit risk prediction using weight-adjusted boosting ensemble method

With the tremendous development of financial institutions, credit risk prediction (CRP) plays an essential role in granting loans to customers and helps them to minimize their loss because credit approval sometimes results in massive financial loss. So extra attention is needed to identify risky customer. Researchers have designed complex CRP models using artificial intelligence (AI) and statistical techniques to support the financial institutions to take correct business decisions. Though there are various statistical and AI methods available, the recent literature shows that the ensemble-based CRP model provides improved prediction results than single classifier system. The small increase in the performance of CRP model could result in a significant improvement in the profit of financial institutions and banks. This work proposes a weight-adjusted boosting ensemble method (WABEM) using rough set (RS)-based feature selection (FS) technique with the balancing and regression-based preprocessing called RS$$\_$$RFS-WABEM. Regression is used to fill missing value in the records to improve the performance of CRP. Three credit datasets (Australia, German and Japanese) are chosen to validate the feasibility and effectiveness of the proposed ensemble method. The trade-off between the uncertainty and imprecise probability of the proposed classifier model is evaluated using the performance measures such as accuracy and area under the curve. Experimental results show that the proposed ensemble method performs better than other base and ensemble classifier methods.

Deep learning framework to forecast electricity demand

The increasing world population and availability of energy hungry smart devices are major reasons for alarmingly high electricity consumption in the current times. So far, various simulation tools, engineering and Artificial Intelligence based methods are being used to perform optimal electricity demand forecasting. While engineering methods use dynamic equations to forecast, the AI-based methods use historical data to predict future demand. However, modeling of nonlinear electricity demand patterns is still underdeveloped for robust solutions as the existing methods are useful only for handling short-term dependencies. Moreover, the existing methods are static in nature because they are purely historical data driven. In this paper, we propose a deep learning based framework to forecast electricity demand by taking care of long-term historical dependencies. Initially, the cluster analysis is performed on the electricity consumption data of all months to generate season based segmented data. Subsequently, load trend characterization is carried out to have a deeper insight of metadata falling into each of the clusters. Further, Long Short Term Memory network multi-input multi-output models are trained to forecast electricity demand based upon the season, day and interval data. In the present work, we have also incorporated the concept of moving window based active learning to improve prediction results. To demonstrate the applicability and effectiveness of the proposed approach, it is applied to the electricity consumption data of Union Territory Chandigarh, India. Performance of the proposed approach is evaluated by comparing the prediction results with Artificial Neural Network, Recurrent Neural Network and Support Vector Regression models.

SA38 - GWAS-BASED MACHINE LEARNING APPROACH TO PREDICT DULOXETINE RESPONSE AND REMISSION IN MAJOR DEPRESSIVE DISORDER

Background Major Depressive Disorder (MDD) is one of the most prevalent psychiatric disorders and is commonly treated with antidepressant drugs. However, large variability is observed in terms of response to antidepressants. GWAS based Machine Learning (ML) models may be useful to predict treatment response, thus improving antidepressant treatment. Methods A sample of 186 MDD patients received treatment with duloxetine for up to 8 weeks were categorized as “responders” based on a Montgomery-Asberg Depression Rating Scale (MADRS) change >50% from baseline; or “remitters” based on a MADRS score of 10 or lower at end point. The initial dataset (N=186) was randomly divided into training and test sets (70% and 30%, or 130 and 56 individuals respectively), resulting in ten different training and test set pairs. A genome-wide logistic regression was performed to identify potentially significant variants related to duloxetine response/remission and extracted the most promising predictors using LASSO regression. Subsequently, Classification-Regression Trees (CRT) and support vector machines (SVM) were applied to construct models, using ten-fold, repeated cross-validation. Results For the response phenotype, LASSO regression suggested twenty SNPs to be entered in prediction models. Calculated models for response achieved accuracy of 63.43% for CRT and 78.93% for SVM. For the remission phenotype, LASSO filtering suggested five SNPs. Best models were characterized by an accuracy of 65.45% with a sensitivity of 69.31% and specificity of 61.43%. Discussion In this study, the potential of using GWAS data to predict duloxetine response was examined using ML models. These models managed to capture a fraction of responders and remitters (i.e. moderate sensitivity), but failed to filter out non-responders and non-remitters (i.e. low specificity). Inclusion of additional non-genetic variables to create integrated models may help improve prediction results.

A Novel Online Dynamic Temporal Context Neural Network Framework for the Prediction of Road Traffic Flow

Traffic flow exhibits different magnitudes of temporal patterns, such as short-term (daily and weekly) and long-term (monthly and yearly). Existing research into road traffic flow prediction has focused on short-term patterns; little research has been done to determine the effect of different long-term patterns on road traffic flow prediction. Providing more temporal contextual information through the use of different temporal data segments could improve prediction results. In this paper, we have investigated different magnitudes of temporal patterns, such as short-term and long-term, through the use of different temporal data segments to understand how contextual temporal data can improve prediction. Furthermore, to learn temporal patterns dynamically, we have proposed a novel online dynamic temporal context neural network framework. The framework uses different temporal data segments as input features, and during online learning, the updating scheme dynamically determines how useful a temporal data segment (short and long-term temporal patterns) is for prediction, and weights it accordingly for use in the regression model. Therefore, the framework can include short-term and relevant long-term patterns in the regression model leading to improved prediction results. We have conducted a thorough experimental evaluation with a real dataset containing daily, weekly, monthly and yearly data segments. The experiment results show that both short and long-term temporal patterns improved prediction accuracy. In addition, the proposed online dynamical framework improved predication results by 10.8% when compared with a deep gated recurrent unit model.

Virtual Screening with Generative Topographic Maps: How Many Maps Are Required?

Universal generative topographic maps (GTMs) provide two-dimensional representations of chemical space selected for their "polypharmacological competence", that is, the ability to simultaneously represent meaningful activity and property landscapes, associated with many distinct targets and properties. Several such GTMs can be generated, each based on a different initial descriptor vector, encoding distinct structural features. While their average polypharmacological competence may indeed be equivalent, they nevertheless significantly diverge with respect to the quality of each property-specific landscape. In this work, we show that distinct universal maps represent complementary and strongly synergistic views of biologically relevant chemical space. Eight universal GTMs were employed as support for predictive classification landscapes, using more than 600 active/inactive ligand series associated with as many targets from the ChEMBL database (v.23). For nine of these targets, it was possible to extract, from the Directory of Useful Decoys (DUD), truly external sets featuring sufficient "actives" and "decoys" not present in the landscape-defining ChEMBL ligand sets. For each such molecule, projected on every class landscape of a particular universal map, a probability of activity was estimated, in analogy to a virtual screening (VS) experiment. Cross-validated (CV) balanced accuracy on landscape-defining ChEMBL data was unable to predict the success of that landscape in VS. Thus, the universal map with best CV results for a given property should not be prioritized as the implicitly best predictor. For a given map, predictions for many DUD compounds are not trustworthy, according to applicability domain considerations. By contrast, simultaneous application of all universal maps, and rating of the likelihood of activity as the mean returned by all applicable maps, significantly improved prediction results. Performance measures in consensus VS using multiple maps were always superior or similar to those of the best individual map.