Prediction Window Research Articles

Explainable machine learning approach to predict extubation in critically ill ventilated patients: a retrospective study in central Taiwan

BackgroundWeaning from mechanical ventilation (MV) is an essential issue in critically ill patients, and we used an explainable machine learning (ML) approach to establish an extubation prediction model.MethodsWe enrolled patients who were admitted to intensive care units during 2015–2019 at Taichung Veterans General Hospital, a referral hospital in central Taiwan. We used five ML models, including extreme gradient boosting (XGBoost), categorical boosting (CatBoost), light gradient boosting machine (LightGBM), random forest (RF) and logistic regression (LR), to establish the extubation prediction model, and the feature window as well as prediction window was 48 h and 24 h, respectively. We further employed feature importance, Shapley additive explanations (SHAP) plot, partial dependence plot (PDP) and local interpretable model-agnostic explanations (LIME) for interpretation of the model at the domain, feature, and individual levels.ResultsWe enrolled 5,940 patients and found the accuracy was comparable among XGBoost, LightGBM, CatBoost and RF, with the area under the receiver operating characteristic curve using XGBoost to predict extubation was 0.921. The calibration and decision curve analysis showed well applicability of models. We also used the SHAP summary plot and PDP plot to demonstrate discriminative points of six key features in predicting extubation. Moreover, we employed LIME and SHAP force plots to show predicted probabilities of extubation and the rationale of the prediction at the individual level.ConclusionsWe developed an extubation prediction model with high accuracy and visualised explanations aligned with clinical workflow, and the model may serve as an autonomous screen tool for timely weaning.

Machine learning-based failure prediction in industrial maintenance: improving performance by sliding window selection

PurposeMachine learning (ML) models are increasingly being used in industrial maintenance to predict system failures. However, less is known about how the time windows for reading data and making predictions affect performance. Therefore, the purpose of this research is to assess the impact of different sliding windows on prediction performance.Design/methodology/approachThe authors conducted a factorial experiment using high dimensional machine data covering two years of operation, taken from a real industrial case for the production of high-precision milled and turned parts. The impacts of different reading and prediction windows were tested for three ML algorithms (random forest, support vector machines and logistic regression) and four metrics (accuracy, precision, recall and F-score).FindingsThe results reveal (1) the critical role of the prediction window contingent upon the application domain, (2) a non-monotonic relationship between the reading window and performance, and (3) how sliding window selection can systematically be used to improve different facets of performance.Originality/valueThe study's findings advance the knowledge of ML-based failure prediction, by highlighting how systematic variation of two important but yet understudied factors contributes to the development of more useful prediction models.

A Model Fusion Method for Online State of Charge and State of Power Co-Estimation of Lithium-Ion Batteries in Electric Vehicles

In this paper, a model fusion method (MFM) is proposed for online state of charge (SOC) and state of power (SOP) co-estimation of lithium-ion batteries (LIBs) in electric vehicles (EVs). Firstly, a particle swarm optimization-genetic algorithm (PSO-GA) method is cooperated with a 2-RC <sub xmlns:mml="http://www.w3.org/1998/Math/MathML" xmlns:xlink="http://www.w3.org/1999/xlink">CPE</sub> fractional-order model (FOM) to construct battery open-circuit voltage (OCV)-SOC curve, which only relies on a part of dynamic load profile without the prior knowledge of an initial SOC. Secondly, a dual extended Kalman filter (DEKF) algorithm based on a 1-RC model is employed to identify the model parameters and estimate battery SOC with the extracted OCV-SOC curve. Furthermore, battery polarization dynamics in a SOP prediction window is analyzed from two aspects: (1) self-recovery; and (2) current excitation. They are separately simulated using 2-RC <sub xmlns:mml="http://www.w3.org/1998/Math/MathML" xmlns:xlink="http://www.w3.org/1999/xlink">CPE</sub> FOM and 1-RC model, and then integrated through a model fusion for online SOP estimation, which enables an analytical expression of battery peak charge/discharge current in a prediction window without weakening the nonlinear characteristic of FOM. Experimental results demonstrate the improved performance of the proposed MFM for online discharge SOP estimation, where the mean absolute error and root mean square error are only 0.288 W and 0.35 W, respectively, under the urban dynamometer driving schedule profile.

A time-aware attention model for prediction of acute kidney injury after pediatric cardiac surgery.

Acute kidney injury (AKI) is a common complication after pediatric cardiac surgery, and the early detection of AKI may allow for timely preventive or therapeutic measures. However, current AKI prediction researches pay less attention to time information among time-series clinical data and model building strategies that meet complex clinical application scenario. This study aims to develop and validate a model for predicting postoperative AKI that operates sequentially over individual time-series clinical data. A retrospective cohort of 3386 pediatric patients extracted from PIC database was used for training, calibrating, and testing purposes. A time-aware deep learning model was developed and evaluated from 3 clinical perspectives that use different data collection windows and prediction windows to answer different AKI prediction questions encountered in clinical practice. We compared our model with existing state-of-the-art models from 3 clinical perspectives using the area under the receiver operating characteristic curve (ROC AUC) and the area under the precision-recall curve (PR AUC). Our proposed model significantly outperformed the existing state-of-the-art models with an improved average performance for any AKI prediction from the 3 evaluation perspectives. This model predicted 91% of all AKI episodes using data collected at 24h after surgery, resulting in a ROC AUC of 0.908 and a PR AUC of 0.898. On average, our model predicted 83% of all AKI episodes that occurred within the different time windows in the 3 evaluation perspectives. The calibration performance of the proposed model was substantially higher than the existing state-of-the-art models. This study showed that a deep learning model can accurately predict postoperative AKI using perioperative time-series data. It has the potential to be integrated into real-time clinical decision support systems to support postoperative care planning.

Applying Machine Learning to Predict Esophageal Cancer Recurrence after Esophagectomy

INTRODUCTION: Artificial intelligence and machine learning (ML) models have recently been adapted in healthcare applications with promising results. The objective of this proof-of-concept study was to develop a ML model designed to predict esophageal cancer recurrence after esophagectomy. METHODS: We conducted a retrospective study of 260 consecutive patients who underwent esophagectomy for esophageal cancer from 2009 through 2018. Over 20,000 patient-specific characteristics were collected. Risk prediction models for different prediction windows were constructed via a sequential forward selection process. Five traditional machine learning algorithms including Logistic Regression (LR), Support Vector Machine (SVM), Random Forest (RF), Decision Tree (DT), and Naive Bayes (NB) were included in our analysis. Model performance was assessed by calculating sensitivity, specificity, positive predictive value (PPV), F1 score, area under the receiver operating characteristic curve (AUC), and overall accuracy using five-fold cross-validation. RESULTS: Of the 260 patients, 121 (46.5%) experienced cancer recurrence at a median of 267 days (49-2,057 days). The highest AUC for prediction of recurrence at any time after esophagectomy was 0.82 (SD ± 0.02) by DT. Similarly, the accuracy of the model to predict whether or not a new subject inserted into the model will have recurrence at any time was 82% (0.82 ± 0.03) by RF. The LR algorithm had the highest accuracy (92%, 0.92 ± 0.01) for predicting recurrence within 180 days after esophagectomy (Table). CONCLUSION: This proof-of-concept study demonstrates the feasibility of using machine learning models to predict esophageal cancer recurrence. The accuracy and AUC of the machine learning models exceeded 80% in all recurrence prediction timeframes. Table. - Overall Performance of the Machine Learning Models Prediction window No. (%) of subjects with recurrence/no. (%) of subjects without recurrence Model Accuracy (SD) AUC (SD) Recurrence at any time after esophagectomy 121 (46.5) / 139 (53.5) LR 0.80 (0.05) 0.81 (0.04) SVM 0.77 (0.05) 0.79 (0.06) RF 0.82 (0.03) 0.82 (0.05) DT 0.79 (0.03) 0.82 (0.02) NB 0.80 (0.04) 0.78 (0.06) Recurrence within 180 days after esophagectomy 37 (14.2) / 223 (85.8) LR 0.92 (0.01) 0.78 (0.05) SVM 0.92 (0.02) 0.81 (0.07) RF 0.90 (0.03) 0.73 (0.10) DT 0.90 (0.02) 0.70 (0.04)

Forecasting the moisture dynamics of a landfill capping system comprising different geosynthetics: A NARX neural network approach

Engineered landfill capping systems consist of geosynthetics and soil layers, which often experience inconsistent and extreme weather events throughout their service life. Complex moisture dynamics in the capping layers can be created by these weather events in combination with other field conditions and can be detrimental to the system's integrity. The limited data on the hydraulic performance of landfill capping systems is a major challenge that hinders the development, validation, and calibration of models that can be used for realistic forecasting of these dynamics. Using the field-level data collected at the Bletchley landfill site, UK, this study develops a data-driven forecasting approach employing a non-linear autoregressive neural network with exogenous inputs (NARX). The data includes precipitation and volumetric water content (VWC) of the capping soil overlaying different geosynthetic layers recorded from Nov 2011 to July 2012. The NARX network was trained using the VWC data as inputs and precipitation data as the exogenous input. Also, the accuracy of NARX predictions was compared against that of a state-space statistical model. NARX-predicted VWC values for a period of 21-days ahead are distributed with a mean error of 0.05 and a standard deviation of 0.2. In the majority of prediction windows, NARX approach outperforms the state-space model. For all NARX prediction periods, RMSEr has been less than 10% for the cuspated core geocomposite. Comparatively, RMSEr values increased to approximately 15% and 19% for the non-woven needle-punched geotextile and the non-woven needle-punched geotextile with band drains, respectively.

S&amp;P 500 Index and Volatility Forecast of Chinese Stock Market

The main purpose of this article is to examine the role of the S&amp;P 500 index in predicting the volatility of China's stock market. Our work is based on the autoregressive model (AR). We further extend this simple benchmark model by adding the volatility of the S&amp;P 500 index. Intrasample regression shows that after adding this indicator, the overall goodness of fit of the model is rising, the explanatory ability is enhanced, and the added variables are also very significant. The out of sample prediction shows that, in terms of statistical test, the extended model has a positive out of sample R-square compared with the benchmark model, and has passed the CW test; In terms of economic test, we find that the extended model has positive CER and Sharp Ratio (SR) compared with the benchmark model. The out of sample predictions of these two aspects show that the newly added S&amp;P 500 index has a good prediction effect. In addition, we also conducted various robustness tests, such as replacing the previous dependent variable (Shanghai Stock Exchange Index) with the CSI 300 index, replacing the previous extended window with a rolling window for prediction, and extending the previous single period prediction to multi period prediction. In the multi period forecast, we found that the S&amp;P 500 index is only effective in a short period, for example, within 3 months, but cannot play a predictive role when it is extended to 6 months.

MACHINE LEARNING MODEL FOR GLARE PREDICTION IN OFFICES WITH SIMPLE ARCHITECTURAL FEATURES

ABSTRACT Daylight glare index (DGI), daylight glare probability (DGP) and glare-sensation (GS) predictive models are the widely used glare indices for the assessment of occupant visual comfort in daylit spaces. This paper presents the development and implementation of Machine Learning models to predict these glare indices. The training and validation data sets were collected from sensors incorporated in the test room with motorized Venetian Blinds and dimmable LED luminaires. Predictor and response data were obtained from conventional sensors, digital cameras, and the EVALGLARE Software. The regression models predict DGI and DGP, whereas the classification model predicts GS. In addition to standard statistical error evaluation metrics, the hypothesis test assesses the performance of regression/classification models. The results reveal that Ensemble Tree (ET) models are highly accurate at predicting glare indices. The proposed technique attempts to simplify the existing traditional Glare Index(GI) estimation method. The combination of real-time daylight glare prediction and suitable window shading control increases occupant visual comfort. A high dynamic image-based system is employed to verify the measurements made using traditional sensors.

Developing Machine Learning Algorithms on Routinely Collected Administrative Health Data - Lessons from Ontario, Canada.

There has been considerable growth in the development of machine learning models for clinical applications; however, less attention has been paid to applications at the health systems level. Here, we survey recent models developed using provincial administrative health data holdings in Ontario, Canada to synthesize key learnings across use cases. We have developed four models in the areas of diabetes incidence and complications, hospitalization due to ambulatory care sensitive conditions, and hospitalization due to SARS-CoV-2 infection. Our team was highly multidisciplinary with expertise across clinical medicine, administrative health data, epidemiology, and computer science. We used a “sliding window” approach to aggregate healthcare events across multiple health administrative data sets chronologically and map them dynamically onto a patient timeline. Tree-based algorithms, specifically gradient boosted decision trees, are well suited for the underlying tabular structure of administrative data and were used for each prediction task. Our models achieved excellent discrimination, measured by the area under the receiver operating characteristic curve, between 0.77-0.85 at prediction windows between 30 days and 3 years in advance. They were also well-calibrated, both in-the-large and in population subgroups such as older adults, those living in rural areas, and the materially deprived. Measures of feature importance revealed that our models were leveraging predictors across administrative datasets (e.g. demographics, interactions with the healthcare system, medications) in intuitive and defensible ways. Finally, we demonstrated the utility of our models with “recall at top k” metrics - for example, the top 1% of patients predicted at risk of diabetes complications represented a cost of over $400 million to the healthcare system. We identify three key learnings needed for the successful application of machine learning methods to health administrative data: synergy between nature of training data and intended algorithm use, adherence to methodological best practices for rigour and transparency, and multidisciplinary teams with expertise across data provenance, methodological approach, and impact assessment.

Passively Captured Interpersonal Social Interactions and Motion From Smartphones for Predicting Decompensation in Heart Failure: Observational Cohort Study.

BackgroundHeart failure (HF) is a major cause of frequent hospitalization and death. Early detection of HF symptoms using smartphone-based monitoring may reduce adverse events in a low-cost, scalable way.ObjectiveWe examined the relationship of HF decompensation events with smartphone-based features derived from passively and actively acquired data.MethodsThis was a prospective cohort study in which we monitored HF participants’ social and movement activities using a smartphone app and followed them for clinical events via phone and chart review and classified the encounters as compensated or decompensated by reviewing the provider notes in detail. We extracted motion, location, and social interaction passive features and self-reported quality of life weekly (active) with the short Kansas City Cardiomyopathy Questionnaire (KCCQ-12) survey. We developed and validated an algorithm for classifying decompensated versus compensated clinical encounters (hospitalizations or clinic visits). We evaluated models based on single modality as well as early and late fusion approaches combining patient-reported outcomes and passive smartphone data. We used Shapley additive explanation values to quantify the contribution and impact of each feature to the model.ResultsWe evaluated 28 participants with a mean age of 67 years (SD 8), among whom 11% (3/28) were female and 46% (13/28) were Black. We identified 62 compensated and 48 decompensated clinical events from 24 and 22 participants, respectively. The highest area under the precision-recall curve (AUCPr) for classifying decompensation was with a late fusion approach combining KCCQ-12, motion, and social contact features using leave-one-subject-out cross-validation for a 2-day prediction window. It had an AUCPr of 0.80, with an area under the receiver operator curve (AUC) of 0.83, a positive predictive value (PPV) of 0.73, a sensitivity of 0.77, and a specificity of 0.88 for a 2-day prediction window. Similarly, the 4-day window model had an AUC of 0.82, an AUCPr of 0.69, a PPV of 0.62, a sensitivity of 0.68, and a specificity of 0.87. Passive social data provided some of the most informative features, with fewer calls of longer duration associating with a higher probability of future HF decompensation.ConclusionsSmartphone-based data that includes both passive monitoring and actively collected surveys may provide important behavioral and functional health information on HF status in advance of clinical visits. This proof-of-concept study, although small, offers important insight into the social and behavioral determinants of health and the feasibility of using smartphone-based monitoring in this population. Our strong results are comparable to those of more active and expensive monitoring approaches, and underscore the need for larger studies to understand the clinical significance of this monitoring method.

Capture and Prediction of Rainfall-Induced Landslide Warning Signals Using an Attention-Based Temporal Convolutional Neural Network and Entropy Weight Methods.

The capture and prediction of rainfall-induced landslide warning signals is the premise for the implementation of landslide warning measures. An attention-fusion entropy weight method (En-Attn) for capturing warning features is proposed. An attention-based temporal convolutional neural network (ATCN) is used to predict the warning signals. Specifically, the sensor data are analyzed using Pearson correlation analysis after obtaining data from the sensors on rainfall, moisture content, displacement, and soil stress. The comprehensive evaluation score is obtained offline using multiple entropy weight methods. Then, the attention mechanism is used to weight and sum different entropy values to obtain the final landslide hazard degree (LHD). The LHD realizes the warning signal capture of the sensor data. The prediction process adopts a model built by ATCN and uses a sliding window for online dynamic prediction. The input is the landslide sensor data at the last moment, and the output is the LHD at the future moment. The effectiveness of the method is verified by two datasets obtained from the rainfall-induced landslide simulation experiment.

Acoustic detection of the wood borer, Semanotus bifasciatus, as an early monitoring technology.

Semanotus bifasciatus Motschulsky (Coleoptera: Cerambycidae) is one of the most destructive wood-boring pests of Platycladus trees in East Asia, threatening the protection of antique cypresses and urban ecological safety. Early identification of Semanotus bifasciatus attacks can help forest managers mitigate the infestation before it turns into an outbreak. Acoustic detection technology is a non-destructive and continuous monitoring method with the potential to early identify and accurately evaluate the wood-boring damage. However, few studies have focused on the detection timing and corresponding acoustic features. In this study, we employed a manipulated insect infestation experiment to identify time windows in which early instar Semanotus bifasciatus larvae are most actively boring and feeding within logs and to identify acoustic features that distinguish larval sounds from typical background noise. The Semanotus bifasciatus larvae produced sounds most frequently between 13:00 and 20:00 while sounds were detectable from the first to the third instar during the larval growth stage, indicating a suitable time window for early detection. The stepwise regression (SR) model was optimal for detecting the larval instar [coefficient of determination (R2 )=0.71, root mean squared error of prediction (RMSEp )=0.42, and relative percent deviation (RPD)= 3.38] while the best model for predicting larval population size was the partial least squares regression (PLSR) model (R2 =0.97, RMSEp =61.96, and RPD=28.87). This study developed an acoustic method for identifying the early attack of Semanotus bifasciatus (including detection time window, feature variables and models for larval instar prediction and population size estimation). This technology integrated with internet of things (IoT) framework can be of value in developing an automated monitoring system for forest wood borer, and provide necessary guidance for integrated pest management (IPM). © 2022 Society of Chemical Industry.

Exploring the Utility of Crutch Force Sensors to Predict User Intent in Assistive Lower Limb Exoskeletons.

The adoption of assistive lower limb exoskeletons in built environments is reliant on the further development of these devices to handle the varied conditions experienced in everyday life. The required development includes more varied and flexible gait patterns, but also appropriate user interfaces to enable fluid gait. This work explores the properties of an algorithm used to predict user intent based on sensors onboard a user-balanced robotic exoskeleton system. Specifically, classification algorithms built with different input data sets are compared - with varying detail of the interaction forces between the crutches and the ground, and the duration of the data sample used to make the prediction. Data were collected with one able-bodied participant using an exoskeleton, training three independent classifiers corresponding to different exoskeleton states. The results indicate the value of including information about the interaction forces between the crutches and the ground in improving prediction accuracy, with increasing prediction window also generally resulting in an increase in prediction accuracy. Whilst no categorical recommendation can be made with respect to either parameter, these results provide a baseline which can be used in conjunction deliberate consideration of the costs associated with implementation.

Falls prediction using the nursing home minimum dataset.

The purpose of the study was to develop and validate a model to predict the risk of experiencing a fall for nursing home residents utilizing data that are electronically available at the more than 15 000 facilities in the United States. The fall prediction model was built and tested using 2 extracts of data (2011 through 2013 and 2016 through 2018) from the Long-term Care Minimum Dataset (MDS) combined with drug data from 5 skilled nursing facilities. The model was created using a hybrid Classification and Regression Tree (CART)-logistic approach. The combined dataset consisted of 3985 residents with mean age of 77 years and 64% female. The model's area under the ROC curve was 0.668 (95% confidence interval: 0.643-0.693) on the validation subsample of the merged data. Inspection of the model showed that antidepressant medications have a significant protective association where the resident has a fall history prior to admission, requires assistance to balance while walking, and some functional range of motion impairment in the lower body; even if the patient exhibits behavioral issues, unstable behaviors, and/or are exposed to multiple psychotropic drugs. The novel hybrid CART-logit algorithm is an advance over the 22 fall risk assessment tools previously evaluated in the nursing home setting because it has a better performance characteristic for the fall prediction window of ≤90 days and it is the only model designed to use features that are easily obtainable at nearly every facility in the United States.

Explainable machine learning for real-time deterioration alert prediction to guide pre-emptive treatment

The Electronic Medical Record (EMR) provides an opportunity to manage patient care efficiently and accurately. This includes clinical decision support tools for the timely identification of adverse events or acute illnesses preceded by deterioration. This paper presents a machine learning-driven tool developed using real-time EMR data for identifying patients at high risk of reaching critical conditions that may demand immediate interventions. This tool provides a pre-emptive solution that can help busy clinicians to prioritize their efforts while evaluating the individual patient risk of deterioration. The tool also provides visualized explanation of the main contributing factors to its decisions, which can guide the choice of intervention. When applied to a test cohort of 18,648 patient records, the tool achieved 100% sensitivity for prediction windows 2–8 h in advance for patients that were identified at 95%, 85% and 70% risk of deterioration.

Remaining useful life prediction of Lithium-ion battery using improved support vector regression and fuzzy information granulation

With the widespread use of lithium-ion batteries in various fields, battery Prognostics and Health Management (PHM) technologies are gaining more and more attention. Repeated use of batteries can lead to degradation of battery performance and thus affect battery life. Accurate prediction of the remaining useful life (RUL) of batteries is crucial and is the most central issue in battery PHM. In this paper, a method based on a combination of fuzzy information granulation (FIG) and support vector regression with artificial bee colony optimization (ABC-SVR) is proposed to estimate the RUL of Lithium-ion batteries. First, the capacity degradation data are divided into several windows using the FIG method. Second, the maximum and minimum values of each window are predicted separately using the ABC-SVR algorithm to obtain the information of the prediction window. Finally, the missing values of the prediction windows are complemented by the linear interpolation method to obtain the complete capacity prediction values, and the remaining useful life of the battery can be calculated according to the failure threshold. The results show that the proposed method obtains the RUL value with high accuracy.

Ensemble Simulations of the 2012 July 12 Coronal Mass Ejection with the Constant-turn Flux Rope Model

Flux-rope-based magnetohydrodynamic modeling of coronal mass ejections (CMEs) is a promising tool for prediction of the CME arrival time and magnetic field at Earth. In this work, we introduce a constant-turn flux rope model and use it to simulate the 2012 July 12 16:48 CME in the inner heliosphere. We constrain the initial parameters of this CME using the graduated cylindrical shell (GCS) model and the reconnected flux in post-eruption arcades. We correctly reproduce all the magnetic field components of the CME at Earth, with an arrival time error of approximately 1 hr. We further estimate the average subjective uncertainties in the GCS fittings by comparing the GCS parameters of 56 CMEs reported in multiple studies and catalogs. We determined that the GCS estimates of the CME latitude, longitude, tilt, and speed have average uncertainties of 5.°74, 11.°23, 24.°71, and 11.4%, respectively. Using these, we have created 77 ensemble members for the 2012 July 12 CME. We found that 55% of our ensemble members correctly reproduce the sign of the magnetic field components at Earth. We also determined that the uncertainties in GCS fitting can widen the CME arrival time prediction window to about 12 hr for the 2012 July 12 CME. On investigating the forecast accuracy introduced by the uncertainties in individual GCS parameters, we conclude that the half-angle and aspect ratio have little impact on the predicted magnetic field of the 2012 July 12 CME, whereas the uncertainties in longitude and tilt can introduce relatively large spread in the magnetic field predicted at Earth.

Customer active power consumption prediction for the next day based on historical profile

Energy consumption prediction application is one of the most important fields that is artificially controlled with Artificial Intelligence technologies to maintain accuracy for electricity market costs reduction. This work presents a way to build and apply a model to each costumer in residential buildings. This model is built by using Long Short Term Memory (LSTM) networks to address a demonstration of time-series prediction problem and Deep Learning to take into consideration the historical consumption of customers and hourly load profiles in order to predict future consumption. Using this model, the most probable sequence of a certain industrial customer’s consumption levels for a coming day is predicted. In the case of residential customers, determining the particular period of the prediction in terms of either a year or a month would be helpful and more accurate due to changes in consumption according to the changes in temperature and weather conditions in general. Both of them are used together in this research work to make a wide or narrow prediction window. A test data set for a set of customers is used. Consumption readings for any customer in the test data set applying LSTM model are varying between minimum and maximum values of active power consumption. These values are always alternating during the day according to customer consumption behavior. This consumption variation leads to leveling all readings to be determined in a finite set and deterministic values. These levels could be then used in building the prediction model. Levels of consumption’s are modeling states in the transition matrix. Twenty five readings are recorded per day on each hour and cover leap years extra ones. Emission matrix is built using twenty five values numbered from one to twenty five and represent the observations. Calculating probabilities of being in each level (node) is also covered. Logistic Regression Algorithm is used to determine the most probable nodes for the next 25 hours in case of residential or industrial customers. Index Terms—Smart Grids, Load Forecasting, Consumption Prediction, Long Short Term Memory (LSTM), Logistic Regression Algorithm, Load Profile, Electrical Consumption.

Flexible-Window Predictions on Electronic Health Records.

Various types of machine learning techniques are available for analyzing electronic health records (EHRs). For predictive tasks, most existing methods either explicitly or implicitly divide these time-series datasets into predetermined observation and prediction windows. Patients have different lengths of medical history and the desired predictions (for purposes such as diagnosis or treatment) are required at different times in the future. In this paper, we propose a method that uses a sequence-to-sequence generator model to transfer an input sequence of EHR data to a sequence of user-defined target labels, providing the end-users with "flexible" observation and prediction windows to define. We use adversarial and semi-supervised approaches in our design, where the sequence-to-sequence model acts as a generator and a discriminator distinguishes between the actual (observed) and generated labels. We evaluate our models through an extensive series of experiments using two large EHR datasets from adult and pediatric populations. In an obesity predicting case study, we show that our model can achieve superior results in flexible-window prediction tasks, after being trained once and even with large missing rates on the input EHR data. Moreover, using a number of attention analysis experiments, we show that the proposed model can effectively learn more relevant features in different prediction tasks.

Holistic autonomous model for early detection of downhole drilling problems in real-time

Due to the recent increase in drilling operations complexity, the frequency of undesirable downhole events occurring while drilling a well is in ascend trend leading to substantial growth in non-productive time. Consequently, overall drilling costs become sky-high, a moment in which earlier and precise detection of the downhole drilling problems becomes a crucial factor in cost reduction. This paper presents an intelligent algorithm that can automatically analyze real-time drilling data and accurately detect and verify the presence of the most common downhole drilling problems upon their effective inception, which allows corrective measures to be applied at the appropriate time, resulting in a reduction of the negative impact and the associated cost of the detected downhole failure. The presented algorithm relies on constructing a risk predictive window by integrating a stochastic model with a data-driven model driven from real-time data of the surface and /or subsurface drilling parameters to detect the downhole problems. The process starts by building predictive models for predefined drilling parameters. Based on the natural distribution of the Residual Errors (REs) obtained while building the productive model, the best probability model that fits the REs data is picked. The statistical properties of the selective probabilistic models and the real-time predictive values of the predefined drilling parameters are used to generate multiple dimensional risk predictive windows; the indicated risk predictive window could have one or two dimensions, depending on the number of pre-built productive models. The downhole drilling problem is detected by observing and comparing the real-time measured value of the used drilling parameters with the risk predictive window; consecutive data points located outside the risk predictive window are considered abnormal. An alarm is triggered when the number of sequent outliers reaches a predetermined boundary. The developed algorithm was tested on a historical drilling dataset in which different downhole incidences occurred. The results show that the algorithm successfully detected all of the events at an average of 120 min before the officially recorded time.