Selective Prediction LSTM for Time Series Health Datasets using Unit-wise Batch Standardization: Algorithm Development and Validation (Preprint)

2021 ◽  
Author(s):  
Baek Hwan Cho ◽  
Borum Nam ◽  
Joo Young Kim ◽  
In Young Kim
Keyword(s):  
Time Series ◽  
Confidence Level ◽  
Structural Characteristics ◽  
False Negative ◽  
Health Data ◽  
Classification Systems ◽  
Superior Performance ◽  
Selection Function ◽  
Learning Stage ◽  
Reject Option

BACKGROUND In any healthcare system, both the classification of data and the confidence level of the classification are important. A selective prediction model is therefore needed to classify time-series health data according to confidence levels of prediction. OBJECTIVE The aim of this study is to develop a method using Long short-term memory (LSTM) models with reject option for time-series health data classification. METHODS To implement a reject option of classification output in LSTM models, an existing selective prediction method was adopted. However, a conventional selection function approach to LSTM does not achieve acceptable performance at the learning stage. To tackle this problem, we propose unit-wise batch standardization (UBS), which attempts to normalize each hidden unit in LSTM to reflect the structural characteristics of LSTM with respect to selection function. RESULTS From the results, the ability of our method to approximate the target confidence level was compared by coverage violations for two time series health datasets consisting of human activity and arrhythmia. For both datasets, our approach yielded lower average coverage violations (0.98% and 1.79% for each dataset) than conventional approach. In addition, the classification performance using the reject option was compared with other normalization methods. Our method demonstrates superior performance with respect to selective risk (12.63% and 17.82% for each dataset), false-positive rates (2.09% and 5.80% for each dataset), and false-negative rates (10.58% and 17.24% for each dataset). CONCLUSIONS We conclude that our normalization approach can help make selective predictions for time-series health data. We expect this technique will give users more confidence in classification systems and improve collaborative efforts between human and artificial intelligence levels in the medical field through the use of classification that reflects confidence.

scholarly journals Marketing Mix Modeling Using PLS-SEM, Bootstrapping the Model Coefficients

Mathematics ◽  
2021 ◽  
Vol 9 (15) ◽  
pp. 1832
Author(s):  
Mariano Méndez-Suárez
Keyword(s):  
Time Series ◽  
Structural Changes ◽  
Marketing Mix ◽  
Structural Equations ◽  
Superior Performance ◽  
Parameter Estimates ◽  
Significance Analysis ◽  
Autocorrelation Structure ◽  
Multichannel Retailing ◽  
Black Friday

Partial least squares structural equations modeling (PLS-SEM) uses sampling bootstrapping to calculate the significance of the model parameter estimates (e.g., path coefficients and outer loadings). However, when data are time series, as in marketing mix modeling, sampling bootstrapping shows inconsistencies that arise because the series has an autocorrelation structure and contains seasonal events, such as Christmas or Black Friday, especially in multichannel retailing, making the significance analysis of the PLS-SEM model unreliable. The alternative proposed in this research uses maximum entropy bootstrapping (meboot), a technique specifically designed for time series, which maintains the autocorrelation structure and preserves the occurrence over time of seasonal events or structural changes that occurred in the original series in the bootstrapped series. The results showed that meboot had superior performance than sampling bootstrapping in terms of the coherence of the bootstrapped data and the quality of the significance analysis.

scholarly journals Enhancing Wind Turbine Power Forecast via Convolutional Neural Network

Electronics ◽  
2021 ◽  
Vol 10 (3) ◽  
pp. 261
Author(s):  
Tianyang Liu ◽  
Zunkai Huang ◽  
Li Tian ◽  
Yongxin Zhu ◽  
Hui Wang ◽  
...  
Keyword(s):  
Time Series ◽  
Wind Speed ◽  
Wind Turbine ◽  
Wind Power ◽  
Rapid Development ◽  
Superior Performance ◽  
Learning Platform ◽  
Novel Approach ◽  
Wind Power Forecast ◽  
Transformation Methods

The rapid development in wind power comes with new technical challenges. Reliable and accurate wind power forecast is of considerable significance to the electricity system’s daily dispatching and production. Traditional forecast methods usually utilize wind speed and turbine parameters as the model inputs. However, they are not sufficient to account for complex weather variability and the various wind turbine features in the real world. Inspired by the excellent performance of convolutional neural networks (CNN) in computer vision, we propose a novel approach to predicting short-term wind power by converting time series into images and exploit a CNN to analyze them. In our approach, we first propose two transformation methods to map wind speed and precipitation data time series into image matrices. After integrating multi-dimensional information and extracting features, we design a novel CNN framework to forecast 24-h wind turbine power. Our method is implemented on the Keras deep learning platform and tested on 10 sets of 3-year wind turbine data from Hangzhou, China. The superior performance of the proposed method is demonstrated through comparisons using state-of-the-art techniques in wind turbine power forecasting.

Optimization of Statistical Methodologies for Anomaly Detection in Gas Turbine Dynamic Time Series

2017 ◽  
Author(s):  
Giuseppe Fabio Ceschini ◽  
Nicolò Gatta ◽  
Mauro Venturini ◽  
Thomas Hubauer ◽  
Alin Murarasu
Keyword(s):  
Time Series ◽  
Gas Turbine ◽  
False Negative ◽  
False Negative Rate ◽  
Simulated Data ◽  
Stationary Time Series ◽  
Moving Window ◽  
Positive Rate ◽  
Dynamic Time ◽  
Window Approach

Statistical parametric methodologies are widely employed in the analysis of time series of gas turbine sensor readings. These methodologies identify outliers as a consequence of excessive deviation from a statistically-based model, derived from available observations. Among parametric techniques, the k-σ methodology demonstrates its effectiveness in the analysis of stationary time series. Furthermore, the simplicity and the clarity of this approach justify its direct application to industry. On the other hand, the k-σ methodology usually proves to be unable to adapt to dynamic time series, since it identifies observations in a transient as outliers. As this limitation is caused by the nature of the methodology itself, two improved approaches are considered in this paper in addition to the standard k-σ methodology. The two proposed methodologies maintain the same rejection rule of the standard k-σ methodology, but differ in the portions of the time series from which statistical parameters (mean and standard deviation) are inferred. The first approach performs statistical inference by considering all observations prior to the current one, which are assumed reliable, plus a forward window containing a specified number of future observations. The second approach proposed in this paper is based on a moving window scheme. Simulated data are used to tune the parameters of the proposed improved methodologies and to prove their effectiveness in adapting to dynamic time series. The moving window approach is found to be the best on simulated data in terms of True Positive Rate (TPR), False Negative Rate (FNR) and False Positive Rate (FPR). Therefore, the performance of the moving window approach is further assessed towards both different simulated scenarios and field data taken on a gas turbine.

scholarly journals Design of Seasonal Adjustment Filter Robust to Variations in the Seasonal Behaviour of Time Series

2017 ◽  
Vol 33 (1) ◽  
pp. 155-186
Author(s):  
Marcela Cohen Martelotte ◽  
Reinaldo Castro Souza ◽  
Eduardo Antônio Barros da Silva
Keyword(s):  
Time Series ◽  
Good Alternative ◽  
Superior Performance ◽  
Seasonal Adjustment ◽  
Artificial Data ◽  
Seasonal Fluctuations ◽  
Seasonal Behaviour ◽  
Macroeconomic Time Series ◽  
The One ◽  
Made In

Abstract Considering that many macroeconomic time series present changing seasonal behaviour, there is a need for filters that are robust to such changes. This article proposes a method to design seasonal filters that address this problem. The design was made in the frequency domain to estimate seasonal fluctuations that are spread around specific bands of frequencies. We assessed the generated filters by applying them to artificial data with known seasonal behaviour based on the ones of the real macroeconomic series, and we compared their performance with the one of X-13A-S. The results have shown that the designed filters have superior performance for series with pronounced moving seasonality, being a good alternative in these cases.

scholarly journals Ordinary kriging as a tool to estimate historical daily streamflow records

2016 ◽  
Vol 20 (7) ◽  
pp. 2721-2735 ◽  
Author(s):  
William H. Farmer
Keyword(s):  
Time Series ◽  
Ordinary Kriging ◽  
Order Approximation ◽  
Superior Performance ◽  
Daily Streamflow ◽  
History Of ◽  
First Order Approximation ◽  
And Behavior ◽  
Kriging Approach ◽  
Daily Streamflow Time Series

Abstract. Efficient and responsible management of water resources relies on accurate streamflow records. However, many watersheds are ungaged, limiting the ability to assess and understand local hydrology. Several tools have been developed to alleviate this data scarcity, but few provide continuous daily streamflow records at individual streamgages within an entire region. Building on the history of hydrologic mapping, ordinary kriging was extended to predict daily streamflow time series on a regional basis. Pooling parameters to estimate a single, time-invariant characterization of spatial semivariance structure is shown to produce accurate reproduction of streamflow. This approach is contrasted with a time-varying series of variograms, representing the temporal evolution and behavior of the spatial semivariance structure. Furthermore, the ordinary kriging approach is shown to produce more accurate time series than more common, single-index hydrologic transfers. A comparison between topological kriging and ordinary kriging is less definitive, showing the ordinary kriging approach to be significantly inferior in terms of Nash–Sutcliffe model efficiencies while maintaining significantly superior performance measured by root mean squared errors. Given the similarity of performance and the computational efficiency of ordinary kriging, it is concluded that ordinary kriging is useful for first-order approximation of daily streamflow time series in ungaged watersheds.

scholarly journals Image Segmentation of the Ventricular Septum in Fetal Cardiac Ultrasound Videos Based on Deep Learning Using Time-Series Information

Biomolecules ◽  
2020 ◽  
Vol 10 (11) ◽  
pp. 1526 ◽  
Author(s):  
Ai Dozen ◽  
Masaaki Komatsu ◽  
Akira Sakai ◽  
Reina Komatsu ◽  
Kanto Shozu ◽  
...  
Keyword(s):  
Machine Learning ◽  
Time Series ◽  
Image Segmentation ◽  
Superior Performance ◽  
Ventricular Septum ◽  
Ultrasound Screening ◽  
Cardiac Ultrasound ◽  
Segmentation Methods ◽  
Detection Of Objects ◽  
Specific Section

Image segmentation is the pixel-by-pixel detection of objects, which is the most challenging but informative in the fundamental tasks of machine learning including image classification and object detection. Pixel-by-pixel segmentation is required to apply machine learning to support fetal cardiac ultrasound screening; we have to detect cardiac substructures precisely which are small and change shapes dynamically with fetal heartbeats, such as the ventricular septum. This task is difficult for general segmentation methods such as DeepLab v3+, and U-net. Hence, here we proposed a novel segmentation method named Cropping-Segmentation-Calibration (CSC) that is specific to the ventricular septum in ultrasound videos in this study. CSC employs the time-series information of videos and specific section information to calibrate the output of U-net. The actual sections of the ventricular septum were annotated in 615 frames from 421 normal fetal cardiac ultrasound videos of 211 pregnant women who were screened. The dataset was assigned a ratio of 2:1, which corresponded to a ratio of the training to test data, and three-fold cross-validation was conducted. The segmentation results of DeepLab v3+, U-net, and CSC were evaluated using the values of the mean intersection over union (mIoU), which were 0.0224, 0.1519, and 0.5543, respectively. The results reveal the superior performance of CSC.

scholarly journals Reducing the Number of Unnecessary Thyroid Biopsies While Improving Diagnostic Accuracy: Toward the “Right” TIRADS

2018 ◽  
Vol 104 (1) ◽  
pp. 95-102 ◽  
Author(s):  
Giorgio Grani ◽  
Livia Lamartina ◽  
Valeria Ascoli ◽  
Daniela Bosco ◽  
Marco Biffoni ◽  
...  
Keyword(s):  
Risk Stratification ◽  
Thyroid Nodules ◽  
False Negative ◽  
False Negative Rate ◽  
Needle Aspiration ◽  
Classification Systems ◽  
American Thyroid Association ◽  
Predictive Values ◽  
Korean Society ◽  
The Right

Abstract Context The prevalence of thyroid nodules in the general population is increasingly high, and at least half of those biopsied prove to be benign. Sonographic risk-stratification systems are being proposed as “rule-out” tests that can identify nodules that do not require fine-needle aspiration (FNA) cytology. Objective To comparatively assess the performances of five internationally endorsed sonographic classification systems [those of the American Thyroid Association, the American Association of Clinical Endocrinologists, the American College of Radiology (ACR), the European Thyroid Association, and the Korean Society of Thyroid Radiology] in identifying nodules whose FNAs can be safely deferred and to estimate their negative predictive values (NPVs). Design Prospective study of thyroid nodules referred for FNA. Setting Single academic referral center. Patients Four hundred seventy-seven patients (358 females, 75.2%); mean (SD) age, 55.9 (13.9) years. Main Outcome Measures Number of biopsies classified as unnecessary, false-negative rate (FNR), sensitivity, specificity, predictive values, and diagnostic ORs for each system. Results Application of the systems’ FNA criteria would have reduced the number of biopsies performed by 17.1% to 53.4%. The ACR Thyroid Imaging Reporting and Data System (TIRADS) allowed the largest reduction (268 of 502) with the lowest FNR (NPV, 97.8%; 95% CI, 95.2% to 99.2%). Except for the Korean Society of Thyroid Radiology TIRADS, all other systems exhibited significant discriminatory performance but produced significantly smaller reductions in the number of procedures. Conclusions Internationally endorsed sonographic risk stratification systems vary widely in their ability to reduce the number of unnecessary thyroid nodule FNAs. The ACR TIRADS outperformed the others, classifying more than half the biopsies as unnecessary with a FNR of 2.2%.

scholarly journals Global & Geographical Mapping and Visualization Method for Personal/Collective Health Data with 5D World Map System

2020 ◽  
Author(s):  
Shiori Sasaki ◽  
Koji Murakami ◽  
Yasushi Kiyoki ◽  
Asako Uraki
Keyword(s):  
Time Series ◽  
Health Monitoring ◽  
Potential Risk ◽  
Global Analysis ◽  
Collaborative Work ◽  
Health Data ◽  
Multimedia Data ◽  
Personal Health ◽  
Single View ◽  
Semantic Correlation

This paper presents a new knowledge base creation method for personal/collective health data with knowledge of preemptive care and potential risk inspection with a global and geographical mapping and visualization functions of 5D World Map System. The final goal of this research project is a realization of a system to analyze the personal health/bio data and potential-risk inspection data and provide a set of appropriate coping strategies and alert with semantic computing technologies. The main feature of 5D World Map System is to provide a platform of collaborative work for users to perform a global analysis for sensing data in a physical space along with the related multimedia data in a cyber space, on a single view of time-series maps based on the spatiotemporal and semantic correlation calculations. In this application, the concrete target data for world-wide evaluation is (1) multi-parameter personal health/bio data such as blood pressure, blood glucose, BMI, uric acid level etc. and daily habit data such as food, smoking, drinking etc., for a health monitoring and (2) time-series multi-parameter collective health/bio data in the national/regional level for global analysis of potential cause of disease. This application realizes a new multidimensional data analysis and knowledge sharing for both a personal and global level health monitoring and disease analysis. The results are able to be analyzed by the time-series difference of the value of each spot, the differences between the values of multiple places in a focused area, and the time-series differences between the values of multiple locations to detect and predict a potential-risk of diseases.

scholarly journals Neural-Network-Based Building Energy Consumption Prediction with Training Data Generation

Processes ◽  
2019 ◽  
Vol 7 (10) ◽  
pp. 731 ◽  
Author(s):  
Sanghyuk Lee ◽  
Jaehoon Cha ◽  
Moon Keun Kim ◽  
Kyeong Soo Kim ◽  
Van Huy Pham ◽  
...  
Keyword(s):  
Neural Network ◽  
Energy Consumption ◽  
False Negative ◽  
Building Energy ◽  
Training Data ◽  
Superior Performance ◽  
Data Generation ◽  
Building Energy Consumption ◽  
Data Set ◽  
Input Variables

The importance of neural network (NN) modelling is evident from its performance benefits in a myriad of applications, where, unlike conventional techniques, NN modeling provides superior performance without relying on complex filtering and/or time-consuming parameter tuning specific to applications and their wider ranges of conditions. In this paper, we employ NN modelling with training data generation based on sensitivity analysis for the prediction of building energy consumption to improve performance and reliability. Unlike our previous work, where insignificant input variables are successively screened out based on their mean impact values (MIVs) during the training process, we use the receiver operating characteristic (ROC) plot to generate reliable data with a conservative or progressive point of view, which overcomes the issue of data insufficiency of the MIV method: By properly setting boundaries for input variables based on the ROC plot and their statistics, instead of completely screening them out as in the MIV-based method, we can generate new training data that maximize true positive and false negative numbers from the partial data set. Then a NN model is constructed and trained with the generated training data using Levenberg–Marquardt back propagation (LM-BP) to perform electricity prediction for commercial buildings. The performance of the proposed data generation methods is compared with that of the MIV method through experiments, whose results show that data generation using successive and cross pattern provides satisfactory performance, following energy consumption trends with good phase. Among the two options in data generation, i.e., successive and two data combination, the successive option shows lower root mean square error (RMSE) than the combination one by around 400~900 kWh (i.e., 30%~75%).

Well Connectivity Analysis with Deep Learning

2021 ◽  
Author(s):  
Yuanjun Li ◽  
Satomi Suzuki ◽  
Roland Horne
Keyword(s):  
Neural Network ◽  
Time Series ◽  
Deep Learning ◽  
False Negative ◽  
Signal Propagation ◽  
Learning Model ◽  
Short Term ◽  
Production Wells ◽  
Interference Tests ◽  
Deep Learning Model

Abstract Knowledge of well connectivity in a reservoir is crucial, especially for early-stage field development and water injection management. However, traditional interference tests can often take several weeks or even longer depending on the distance between wells and the hydraulic diffusivity of the reservoir. Therefore, instead of physically shutting in production wells, we can take advantage of deep learning methods to perform virtual interference tests. In this study, we first used the historical field data to train the deep learning model, a modified Long- and Short-term Time-series network (LSTNet). This model combines the Convolution Neural Network (CNN) to extract short-term local dependency patterns, the Recurrent Neural Network (RNN) to discover long-term patterns for time series trends, and a traditional autoregressive model to alleviate the scale insensitive problem. To address the time-lag issue in signal propagation, we employed a skip-recurrent structure that extends the existing RNN structure by connecting a current state with a previous state when the flow rate signal from an adjacent well starts to impact the observation well. In addition, we found that wells connected to the same manifold usually have similar liquid production patterns, which can lead to false causation of subsurface pressure communication. Thus we enhanced the model performance by using external feature differences to remove the surface connection in the data, thereby reducing input similarity. This enhancement can also amplify the weak signal and thus distinguish input signals. To examine the deep learning model, we used the datasets generated from Norne Field with two different geological settings: sealing and nonsealing cases. The production wells are placed at two sides of the fault to test the false-negative prediction. With these improvements and with parameter tuning, the modified LSTNet model could successfully indicate the well connectivity for the nonsealing cases and reveal the sealing structures in the sealing cases based on the historical data. The deep learning method we employed in this work can predict well pressure without using hand-crafted features, which are usually formed based on flow patterns and geological settings. Thus, this method should be applicable to general cases and more intuitive. Furthermore, this virtual interference test with a deep learning framework can avoid production loss.

Sign in / Sign up

Export Citation Format

Share Document