Predictive Machine Research Articles

Geochemistry and Multiomics Data Differentiate Streams in Pennsylvania Based on Unconventional Oil and Gas Activity.

ABSTRACTUnconventional oil and gas (UOG) extraction is increasing exponentially around the world, as new technological advances have provided cost-effective methods to extract hard-to-reach hydrocarbons. While UOG has increased the energy output of some countries, past research indicates potential impacts in nearby stream ecosystems as measured by geochemical and microbial markers. Here, we utilized a robust data set that combines 16S rRNA gene amplicon sequencing (DNA), metatranscriptomics (RNA), geochemistry, and trace element analyses to establish the impact of UOG activity in 21 sites in northern Pennsylvania. These data were also used to design predictive machine learning models to determine the UOG impact on streams. We identified multiple biomarkers of UOG activity and contributors of antimicrobial resistance within the order Burkholderiales. Furthermore, we identified expressed antimicrobial resistance genes, land coverage, geochemistry, and specific microbes as strong predictors of UOG status. Of the predictive models constructed (n = 30), 15 had accuracies higher than expected by chance and area under the curve values above 0.70. The supervised random forest models with the highest accuracy were constructed with 16S rRNA gene profiles, metatranscriptomics active microbial composition, metatranscriptomics active antimicrobial resistance genes, land coverage, and geochemistry (n = 23). The models identified the most important features within those data sets for classifying UOG status. These findings identified specific shifts in gene presence and expression, as well as geochemical measures, that can be used to build robust models to identify impacts of UOG development.IMPORTANCE The environmental implications of unconventional oil and gas extraction are only recently starting to be systematically recorded. Our research shows the utility of microbial communities paired with geochemical markers to build strong predictive random forest models of unconventional oil and gas activity and the identification of key biomarkers. Microbial communities, their transcribed genes, and key biomarkers can be used as sentinels of environmental changes. Slight changes in microbial function and composition can be detected before chemical markers of contamination. Potential contamination, specifically from biocides, is especially concerning due to its potential to promote antibiotic resistance in the environment. Additionally, as microbial communities facilitate the bulk of nutrient cycling in the environment, small changes may have long-term repercussions. Supervised random forest models can be used to identify changes in those communities, greatly enhance our understanding of what such impacts entail, and inform environmental management decisions.

Neighborhood-based inference and restricted Boltzmann machine for microbe and drug associations prediction.

BackgroundEfficient identification of microbe-drug associations is critical for drug development and solving problem of antimicrobial resistance. Traditional wet-lab method requires a lot of money and labor in identifying potential microbe-drug associations. With development of machine learning and publication of large amounts of biological data, computational methods become feasible.MethodsIn this article, we proposed a computational model of neighborhood-based inference (NI) and restricted Boltzmann machine (RBM) to predict potential microbe-drug association (NIRBMMDA) by using integrated microbe similarity, integrated drug similarity and known microbe-drug associations. First, NI was used to obtain a score matrix of potential microbe-drug associations by using different thresholds to find similar neighbors for drug or microbe. Second, RBM was employed to obtain another score matrix of potential microbe-drug associations based on contrastive divergence algorithm and sigmoid function. Because generalization ability of individual method is poor, we used an ensemble learning to integrate two score matrices for predicting potential microbe-drug associations more accurately. In particular, NI can fully utilize similar (neighbor) information of drug or microbe and RBM can learn potential probability distribution hid in known microbe-drug associations. Moreover, ensemble learning was used to integrate individual predictor for obtaining a stronger predictor.ResultsIn global leave-one-out cross validation (LOOCV), NIRBMMDA gained the area under the receiver operating characteristics curve (AUC) of 0.8666, 0.9413 and 0.9557 for datasets of DrugVirus, MDAD and aBiofilm, respectively. In local LOOCV, AUCs of 0.8512, 0.9204 and 0.9414 were obtained for NIRBMMDA based on datasets of DrugVirus, MDAD and aBiofilm, respectively. For five-fold cross validation, NIRBMMDA acquired AUC and standard deviation of 0.8569 ± −0.0027, 0.9248 ± −0.0014 and 0.9369 ± −0.0020 on the basis of datasets of DrugVirus, MDAD and aBiofilm, respectively. Moreover, case study for severe acute respiratory syndrome coronavirus 2 (SARS-CoV-2) showed that 13 out of the top 20 predicted drugs were verified by searching literature. The other two case studies indicated that 17 and 17 out of the top 20 predicted microbes for the drug of ciprofloxacin and minocycline were confirmed by identifying published literature, respectively.

A systematic approach to brain dynamics: cognitive evolution theory of consciousness.

The brain integrates volition, cognition, and consciousness seamlessly over three hierarchical (scale-dependent) levels of neural activity for their emergence: a causal or 'hard' level, a computational (unconscious) or 'soft' level, and a phenomenal (conscious) or 'psyche' levelrespectively. The cognitive evolution theory (CET) is based on three general prerequisites: physicalism, dynamism, and emergentism, which entail five consequences about the nature of consciousness: discreteness, passivity, uniqueness, integrity, and graduation. CET starts from the assumption that brains should have primarily evolved as volitional subsystems of organisms, not as prediction machines. This emphasizes the dynamical nature of consciousness in terms of critical dynamics to account for metastability, avalanches, and self-organized criticality of brain processes, then coupling it with volition and cognition in a framework unified over the levels. Consciousness emerges near critical points, and unfolds as a discrete stream of momentary states, each volitionally driven from oldest subcortical arousal systems. The stream is the brain's way of making a difference via predictive (Bayesian) processing. Its objective observables could be complexity measures reflecting levels of consciousness and its dynamical coherency to reveal how much knowledge (information gain) the brain acquires over the stream. CET also proposes a quantitative classification of both disorders of consciousness and mental disorders within that unified framework.

Predictive Machine Learning Model for Bottom Hole Flowing and Average Formation Pressure in Underground Gas Storage

Abstract Considering that injection–production of underground gas storage (UGS) is characterized by periodic and dramatic change, effective and fast model for predicting the pressure of UGS would not only be a valuable tool to figure out pressure variety but also of great benefit in optimizing injection and production. This study proposes a practical pressure prediction procedure for UGS to adapt the imbalances between injection and production on a timely basis. In this work, a first step in establishing a novel correlativity measure algorithm to screen out the objective injector–producer wells is proposed. A continuous feature selection strategy aims at selecting and filtrating feature to form the input variables of the pressure predictive model. Eventually, the long-short term memory model is used to fit the variation of pressure. Besides, an in-depth discussion illustrates the importance of well site division and model sensitivity analysis. The predictive capability of the proposed approach is verified by a real application scenario. Experimental results reveal that predictive relative error is less than 5%, which proves that the above procedure exhibits better prediction performance. The novelty of this work is that it is a purely data-driven approach that can directly interpret conventional surface measurements into intuitive subsurface pressure parameters, ideal for field applications of UGS.

Deep architecture for silica forecasting of a real industrial froth flotation process

Froth Flotation is one of the fundamental processes used in iron ore attainment. It involves various physical–chemical reactions, and its multivariate control is not trivial. Traditionally, plant operators get quality measurements from laboratory analysis of sampled material, which can take hours to be published. In this paper, we present alternatives that eliminate both the necessity to periodically sample material as well as the delay associated with laboratory analysis: a predictive machine learning model working as a soft sensor. Two different classes of models were developed, exploring both deep and shallow approaches of machine learning, with a deep neural network architecture constructed for the particular task at hand. All models yielded accurate predictions, with results favoring the deep neural network when a larger volume of data is available.

Data-Driven Analysis of Hole-Transporting Materials for Perovskite Solar Cells Performance.

We have created a dataset of 269 perovskite solar cells, containing information about their perovskite family, cell architecture, and multiple hole-transporting materials features, including fingerprints, additives, and structural and electronic features. We propose a predictive machine learning model that is trained on these data and can be used to screen possible candidate hole-transporting materials. Our approach allows us to predict the performance of perovskite solar cells with reasonable accuracy and is able to successfully identify most of the top-performing and lowest-performing hole-transporting materials in the dataset. We discuss the effect of data biases on the distribution of perovskite families/architectures on the model’s accuracy and offer an analysis with a subset of the data to accurately study the effect of the hole-transporting material on the solar cell performance. Finally, we discuss some chemical fragments, like arylamine and aryloxy groups, which present a relatively large positive correlation with the efficiency of the cell, whereas other groups, like thiophene groups, display a negative correlation with power conversion efficiency (PCE).

Predicting the Rheological Properties of Super-Plasticized Concrete Using Modeling Techniques.

Interface yield stress (YS) and plastic viscosity (PV) have a significant impact on the pumpability of concrete mixes. This study is based on the application of predictive machine learning (PML) techniques to forecast the rheological properties of fresh concrete. The artificial neural network (NN) and random forest (R-F) PML approaches were introduced to anticipate the PV and YS of concrete. In comparison, the R-F model outperforms the NN model by giving the coefficient of determination (R2) values equal to 0.92 and 0.96 for PV and YS, respectively. In contrast, the model’s legitimacy was also verified by applying statistical checks and a k-fold cross validation approach. The mean absolute error, mean square error, and root mean square error values for R-F models by investigating the YS were noted as 30.36 Pa, 1141.76 Pa, and 33.79 Pa, respectively. Similarly, for the PV, these values were noted as 3.52 Pa·s, 16.48 Pa·s, and 4.06 Pa·s, respectively. However, by comparing these values with the NN’s model, they were found to be higher, which also gives confirmation of R-F’s high precision in terms of predicting the outcomes. A validation approach known as k-fold cross validation was also introduced to authenticate the precision of employed models. Moreover, the influence of the input parameters was also investigated with regard to predictions of PV and YS. The proposed study will be beneficial for the researchers and construction industries in terms of saving time, effort, and cost of a project.

Big Data ETL Process and Its Impact on Text Mining Analysis for Employees’ Reviews

Big data analysis is challenging in the current context for enterprises that would like to apply these capabilities in the human resource sector. This paper will show how an organization can take advantage of the current or former employees’ reviews that are provided on a constant basis on different sites, so that the management can adjust or change business decisions based on employees’ wishes, dissatisfaction or needs. Considering the previously mentioned challenge on big data analysis, this research will first provide the best practice for the collection and transformation of the data proposed for analysis. The second part of this paper presents the extraction of two datasets containing employee reviews using data scraping techniques, the analysis of data by using text mining techniques to retrieve business insights and the comparison of the results for these algorithms. Experimental results with Naïve Bayes, Logistic Regression, K-Nearest Neighbor and Support Vector Machine for employee sentiment prediction showed much better performances for Logistic Regression. Three out of the four analyzed algorithms performed better for the second, triple-size dataset. The final aim of the paper is to provide an end-to-end solution with high performance and reduced costs.

Applying machine learning and predictive modeling to retention and viral suppression in South African HIV treatment cohorts

HIV treatment programs face challenges in identifying patients at risk for loss-to-follow-up and uncontrolled viremia. We applied predictive machine learning algorithms to anonymised, patient-level HIV programmatic data from two districts in South Africa, 2016–2018. We developed patient risk scores for two outcomes: (1) visit attendance ≤ 28 days of the next scheduled clinic visit and (2) suppression of the next HIV viral load (VL). Demographic, clinical, behavioral and laboratory data were investigated in multiple models as predictor variables of attending the next scheduled visit and VL results at the next test. Three classification algorithms (logistical regression, random forest and AdaBoost) were evaluated for building predictive models. Data were randomly sampled on a 70/30 split into a training and test set. The training set included a balanced set of positive and negative examples from which the classification algorithm could learn. The predictor variable data from the unseen test set were given to the model, and each predicted outcome was scored against known outcomes. Finally, we estimated performance metrics for each model in terms of sensitivity, specificity, positive and negative predictive value and area under the curve (AUC). In total, 445,636 patients were included in the retention model and 363,977 in the VL model. The predictive metric (AUC) ranged from 0.69 for attendance at the next scheduled visit to 0.76 for VL suppression, suggesting that the model correctly classified whether a scheduled visit would be attended in 2 of 3 patients and whether the VL result at the next test would be suppressed in approximately 3 of 4 patients. Variables that were important predictors of both outcomes included prior late visits, number of prior VL tests, time since their last visit, number of visits on their current regimen, age, and treatment duration. For retention, the number of visits at the current facility and the details of the next appointment date were also predictors, while for VL suppression, other predictors included the range of the previous VL value. Machine learning can identify HIV patients at risk for disengagement and unsuppressed VL. Predictive modeling can improve the targeting of interventions through differentiated models of care before patients disengage from treatment programmes, increasing cost-effectiveness and improving patient outcomes.

Can machine learning predict pharmacotherapy outcomes? An application study in osteoporosis

Background and objectiveThe specific aim of this study is to develop machine learning models as a clinical approach for personalized treatment of osteoporosis. The model performance on outcome prediction was compared between four machine learning algorithms. MethodsRetrospective, electronic clinical data for patients with suspected or confirmed osteoporosis treated at Wan Fang Hospital between 2011 to 2018 were used as inputs for building the following predictive machine learning models,i.e., artificial neural network (ANN), random forest (RF), support vector machine (SVM) and logistic regression (LR) models. The predicted outcome was defined as an increase/decrease in T-score after treatment. A genetic algorithm was employed to select relevant variables as input features for each model; the leave-one-out method was applied for model building and internal validation. The model with best performance was selected by a separate set of testing. Area under the receiver operating characteristic curve, accuracy, precision, sensitivity and F1 score were calculated to evaluate model performance. Main analysis for all the patients with subclinical or confirmed osteoporosis and subgroup analysis for the patients with confirmed osteoporosis (T score < -2.5) were carried out in this study. ResultsA genetic algorithm was employed to select 12 to 18 features from all 33 variables for the four models. No difference was found in accuracy (ANN, 71.7%; LR, 70.0%; RF, 75.0%; SVM, 66.7%), precision (ANN, 80.0%; LR, 59.3%; RF, 70.0%; SVM, 63.6%), and AUC (ANN, 0.709; LR, 0.731; RF, 0.719; SVM, 0.702) among the ANN, LR, RF and SVM models. Main analysis in performance revealed significant recall in the LR model, as compared to ANN and SVM model; while subgroup revealed significant recall in ANN model, compared to LR and SVM model. ConclusionsMachine learning-based models hold potential in forecasting the outcomes of treatment for osteoporosis via early initiation of first-line therapy for patients with subclinical disease; or a switch to second-line treatment for patients with a high risk of impending treatment failure. This convenient approach can assist clinicians in adjusting treatment tailored to individual patient for prevention of disease progression or ineffective therapy.

From Immunohistochemistry to New Digital Ecosystems: A State-of-the-Art Biomarker Review for Precision Breast Cancer Medicine.

Simple SummaryIn this state-of-the-art breast biomarker review, we have tried to imagine and illustrate future, emerging digital breast cancer ecosystems which allow for greater incorporation of traditional immunohistochemical and molecular biomarkers, WSI, and radiomic features.Breast cancers represent complex ecosystem-like networks of malignant cells and their associated microenvironment. Estrogen receptor (ER), progesterone receptor (PR), and human epidermal growth factor receptor 2 (HER2) are biomarkers ubiquitous to clinical practice in evaluating prognosis and predicting response to therapy. Recent feats in breast cancer have led to a new digital era, and advanced clinical trials have resulted in a growing number of personalized therapies with corresponding biomarkers. In this state-of-the-art review, we included the latest 10-year updated recommendations for ER, PR, and HER2, along with the most salient information on tumor-infiltrating lymphocytes (TILs), Ki-67, PD-L1, and several prognostic/predictive biomarkers at genomic, transcriptomic, and proteomic levels recently developed for selection and optimization of breast cancer treatment. Looking forward, the multi-omic landscape of the tumor ecosystem could be integrated with computational findings from whole slide images and radiomics in predictive machine learning (ML) models. These are new digital ecosystems on the road to precision breast cancer medicine.

Environmental model using life cycle assessment and artificial intelligence techniques to predict impacts on industrial water treatment.

With the advancement of globalization, the growing demand for environmental resources in industrial processes and the increasing availability of data and information, the need to align data modeling concepts with Artificial Intelligence (AI) techniques and existing environmental tools has emerged. From a sustainability perspective, life cycle assessment (LCA) is an extremely important tool in ensuring adequate practices in environmental thinking. It is through the life cycle assessment (LCA) that it is possible to measure the environmental impacts from products and processes, as well as to make projections that minimize these impacts. This research employed an artificial intelligence (AI) methods, namely adaptive neurofuzzy inference system (ANFIS) model, to predict life cycle environmental impacts in industrial water treatment using aluminum sulfate and Tannin-Base biocoagulant. The results show that different AI algorithms are used to build LCA models. The AI algorithms in the studies work from problem identification to the solution stage, so the integration between AI and LCA makes it possible to build predictive machine learning models to enable assertiveness in decision making.

Poster 215: Application of Machine Learning for Predicting Opioid Use After Arthroscopic Meniscal Surgery

Objectives:To identify predictive factors for continued opioid prescriptions after arthroscopic meniscal resection or repair and develop a predictive machine learning model.Methods:Patients undergoing arthroscopic meniscal surgery between August 2013 and February 2017 at a single institution were retrospectively identified. Patient demographic variables were recorded including age, sex, body mass index, and history of chronic opioid usage (> 1 month). Procedural details were recorded such as concomitant procedures, primary versus revision, and whether a partial debridement or a repair was performed. Intraoperative arthritis severity was measured using the Outerbridge Classification. Types of opioid medications prescribed and in which months were documented. For primary analysis, we used a multivariate Cox-Regression model. We then created a naïve Bayesian model, a machine learning classifier that utilizes Bayes’ theorem with an assumption of independence between the variables collected. The model randomly selected 70% of the sample to be trained on while 30% were tested.Results:A total of 735 patient were reviewed. Postoperative opioid refills occurred in 98 patients (16.9%). Using multivariate logistic modeling, independent risk factors for opioid refills included Male sex, larger BMI, chronic preoperative opioid use while meniscus resection demonstrated decreased likelihood of refills. Concomitant procedures, revisions, and presence of arthritis graded by the Outerbridge classification were not significant predictors of opioid refills. The Naïve Bayesian model for extended postoperative opioid use demonstrated good fit with our cohort with an area under the curve of 0.79, sensitivity of 94.5%, PPV of 83%, and a detection rate of 78.2%.Conclusions:After arthroscopic meniscus surgery, preoperative opioid consumption had the strongest association with sustained opioid use >1 month. Intraoperative arthritis was not an independent risk factor for continued refills. A novel machine learning algorithm performed with high accuracy and predictive ability to identify patients filling additional narcotic prescriptions after surgery.Figure 1.Cumulative hazard curve for the risk of using opioids up until 1 year postoperatively.Table 1.Multivariate logistic regression model for postoperative opioid refills.

Rapid Discrimination and Prediction of Ginsengs from Three Origins Based on UHPLC-Q-TOF-MS Combined with SVM.

Ginseng, which contains abundant ginsenosides, grows mainly in the Jilin, Liaoning, and Heilongjiang in China. It has been reported that the quality and traits of ginsengs from different origins were greatly different. To date, the accurate prediction of the origins of ginseng samples is still a challenge. Here, we integrated ultra-high-performance liquid chromatography quadrupole time-of-flight mass spectrometry (UHPLC-Q-TOF-MS) with a support vector machine (SVM) for rapid discrimination and prediction of ginseng from the three main regions where it is cultivated in China. Firstly, we develop a stable and reliable UHPLC-Q-TOF-MS method to obtain robust information for 31 batches of ginseng samples after reasonable optimization. Subsequently, a rapid pre-processing method was established for the rapid screening and identification of 69 characteristic ginsenosides in 31 batches ginseng samples from three different origins. The SVM model successfully distinguished ginseng origin, and the accuracy of SVM model was improved from 83% to 100% by optimizing the normalization method. Six crucial quality markers for different origins of ginseng were screened using a permutation importance algorithm in the SVM model. In addition, in order to validate the method, eight batches of test samples were used to predict the regions of cultivation of ginseng using the SVM model based on the six selected quality markers. As a result, the proposed strategy was suitable for the discrimination and prediction of the origin of ginseng samples.

A Result Analysis of Supervised Machine Learning Approach to Detect Anomaly from Network Traffic

Supervised Machine Learning (SML) is the quest for algorithms that reason from externally given cases to develop general hypotheses, which subsequently make predictions about future instances. Supervised categorization is one of the jobs most commonly carried out by the intelligent systems. This article presents numerous Supervised Machine Learning (ML) classification strategies, evaluates various supervised learning algorithms as well as finds the most effective classification algorithm depending on the data set, the number of instances and variables (features) (features). Seven alternative machine learning methods were considered: Decision Table, Random Forest (RF) , Naïve Bayes (NB) , Support Vector Machine (SVM), utilizing Waikato Environment for Knowledge Analysis (WEKA)machine learning program. To develop the algorithms, Diabetes data set was utilized for the classification with 786 cases with eight attributes as independent variable and one as dependent variable for the analysis. The findings suggest that SVM was determined to be the method with maximum precision and accuracy. Naïve Bayes and Random Forest classification algorithms were shown to be the next accurate after SVM appropriately. The research demonstrates that time spent to create a model and precision (accuracy) is a factor on one hand; while kappa statistic and Mean Absolute Error (MAE) is another element on the other side. Therefore, ML techniques demands precision, accuracy and least error to have supervised predictive machine learning.

P-178 Time to hatching: the most predictive morphokinetic parameter for embryo implantation in machine learning algorithms following time-lapse incubation and blastocyst transfer in oocyte-donation programs

Abstract Study question Which is the most predictive parameter when machine learning (ML) is applied to a known implantation database (KID) of day 5 embryo transfer database in an egg donation program? Summary answer Time to hatching (tiHB) is the most predictive embryonic parameter when machine learning algorithms were used on reproductive data in an oocyte donation program. What is known already Artificial intelligence is becoming an encouraging tool in medicine, also in ART, where the amount of data generated in the IVF lab has dramatically increase, favored by time-lapse technology. Numerous embryo selections algorithms based on logistic regressions have been developed for predicting blastocyst formation and implantation potential, but with machine learning, we can train algorithms and connect different morphological and morphokinetic embryo parameters with implantation or even live birth embryo potential. The aim of this study was to test machine learning algorithms and to identify predictive embryonic morphokinetic parameters when comparing the different models generated after machine learning analysis. Study design, size, duration Retrospective analysis of 405 embryos in a KID obtained after 392 embryo-transfers (13 double and 379 single-ET) performed in an oocyte donation program in 4 fertility clinics (year 2021). Recipientś average age: 42.2±4.2 years. The embryos were cultured in Global® Total® culture medium in Geri® (Genea Biomedx) time-lapse incubators after ICSI until embryo transfer at blastocyst stage. Only sperm samples &amp;gt;1x106 spermatozoa/ml were included. All parameters were registered by one single trained senior embryologist. Participants/materials, setting, methods Thirty-five variables were initially analyzed: classic morphokinetic markers, time intervals (including total thinning time before hatching: tiHB-tFB and total blastulation time before hatching: tiHB-Tcav) and morphological measurements (blastocyst and inner cell mass diameter 110h post-injection). Eighty percent of the data was used for model training and 20% was reserved for model validation. Twelve supervised and unsupervised predictive machine learning models were developed. The software used to carry out the analysis was SPSS (v20.0) R (4.0.5). Main results and the role of chance The basic characteristics of the embryo population were similar. From the 405 embryos transferred, 216 blastocysts came from vitrified oocytes (53.3%). The implantation rate was 57.03% (231 gestational sacs) and the miscarriage rate was 16.8%. The classification-supervised algorithms applied included binary logistic regression, neural networks, support vector machines, neighborhood-based methods, classification trees, boosting and bagging methods. The algorithms were optimized by minimizing the AUC. Cluster analysis (unsupervised) was also performed. In the 6 best predictive models, the variables with the highest relevance were tiHB (hatching initiation time), tiHB-tFB and tiHB-tcav: variables related to hatching initiation. Furthermore, in the cluster analysis, these three variables appeared grouped in the same cluster. From the blastocysts population that implanted, 57.6% (133/231) were initiating hatching, while from those embryos that did not, only 42.3% (74/174) began the hatching process. Other variables such as the diameter of the transferred blastocyst, which we assumed to be valuable as an objective morphological parameter, did not show a high predictive capacity in the models obtained. Blastocyst average diameter of implanting blastocysts was 157.9±24.9 µm and non-implanting was 153.9±26.1 µm. Limitations, reasons for caution Morphology and morphokinetic parameters require subjective annotation and thus might have intrinsic intra-reader variability. Our findings need to be validated prospectively. Wider implications of the findings Time to blastocyst hatching appears to have significative impact in most ML predictive models. Hatching-related variables seems to have predictive power. Despite numerous variables influencing IVF outcome (intrinsic and extrinsic to embryo development) ML and AI approaches may improve the prioritization of the most viable embryo favoring single embryo transfer. Trial registration number 2021ibmad001

Convolutional Neural Network and Support Vector Machine for Prediction of Damage Intensity to Multi-Storey Prefabricated RC Buildings

This paper presents the results of a comparative analysis of Convolutional Neural Network (CNN) and Support Vector Machine (SVM) models created for the prediction of the extent and intensity of damage caused to multi-storey reinforced concrete (RC) buildings. The research was conducted on a group of residential buildings, which were subjected to mining impacts in the form of surface deformations and rock mass tremors during their technical life cycle. Damage to buildings poses a significant threat to the safety of the structure and the serviceability of the buildings. They are often the cause of breaks in thermal insulation, which leads to excessive consumption of thermal energy used for space heating, which in turn contributes to over-emissions of CO2 into the atmosphere. Therefore, this problem is important, not only from a technical dimension, but also includes social, economic, and environmental aspects, which allows it to be classified as an issue of sustainable development in the building industry. As a result of the conducted analysis, among the CNN models, the highest level of classification accuracy was the model obtained using the ADAM (derived from adaptive moment estimation) algorithm, which was also characterized by a very high level of generalization, obtaining 80.35% correctly classified patterns for the training set and 80.52% for the test set. However, its accuracy level was slightly lower than that of the SVM model (85.15% for the training set and 84.42% for the test set), in which Bayesian optimization was used to determine the parameters. The analysis confirmed the effectiveness of the adopted methodology for predicting the extent and intensity of damage. The developed tool can support the optimization of building maintenance management, resulting in reduced economic and environmental expenditures for renovations.

Solving the class imbalance problem using a counterfactual method for data augmentation

Learning from class imbalanced datasets poses challenges for many machine learning algorithms. Many real-world domains are, by definition, class imbalanced by virtue of having a majority class that naturally has many more instances than its minority class (e.g., genuine bank transactions occur much more often than fraudulent ones). Many methods have been proposed to solve the class imbalance problem, among the most popular being oversampling techniques (such as SMOTE). These methods generate synthetic instances in the minority class, to balance the dataset, performing data augmentations that improve the performance of predictive machine learning (ML). In this paper, we advance a novel, data augmentation method (adapted from eXplainable AI), that generates synthetic, counterfactual instances in the minority class. Unlike other oversampling techniques, this method adaptively combines existing instances from the dataset, using actual feature-values rather than interpolating values between instances. Several experiments using four different classifiers and 25 datasets involving binary classes are reported, which show that this Counterfactual Augmentation (CFA) method generates useful synthetic datapoints in the minority class. The experiments also show that CFA is competitive with many other oversampling methods, many of which are variants of SMOTE. The basis for CFA’s performance is discussed, along with the conditions under which it is likely to perform better or worse in future tests.

Events prediction after treatment in HPV-driven oropharyngeal carcinoma using machine learning

ObjectivesOur objective was to develop a predictive model using a machine learning signature to identify patients at high risk of relapse or death after treatment for HPV-positive oropharyngeal carcinoma. Materials and methodsPre-treatment variables of 450 patients with HPV-positive oropharyngeal carcinoma treated with a curative intent comprised clinical items, imaging parameters and histological findings. The events considered were progression or residual disease after treatment, the recurrent disease after a disease-free interval and death. The endpoints were the prediction of events and progression-free survival. After feature Z-score normalisation and selection, random forest classifier models were trained. The best models were evaluated on recall, the F-score, and the ROC AUC metric. The clinical relevance of the best prediction model was evaluated using Kaplan–Meier analysis with a log-rank test. ResultsThe best random forest model predicted the 5-year risk of relapse-free survival with a recall of 79.1%, an F1-score of 81.08%, and an AUC of the ROC curve of 0.89. The models performed poorly for the prediction of specific events of progression only, recurrence only or death only. The clinical relevance of the model was validated with a 5-year relapse-free survival of high-risk patients versus low-risk patients of 23.5% and 80%, respectively (p < 0.0001). ConclusionPatients with HPV-driven oropharyngeal carcinoma at high risk of relapse-free survival could be identified with a predictive machine learning model using patient data before treatment.

Role of assortativity in predicting burst synchronization using echo state network.

In this study, we use a reservoir computing based echo state network (ESN) to predict the collective burst synchronization of neurons. Specifically, we investigate the ability of ESN in predicting the burst synchronization of an ensemble of Rulkov neurons placed on a scale-free network. We have shown that a limited number of nodal dynamics used as input in the machine can capture the real trend of burst synchronization in this network. Further, we investigate the proper selection of nodal inputs of degree-degree (positive and negative) correlated networks. We show that for a disassortative network, selection of different input nodes based on degree has no significant role in the machine's prediction. However, in the case of assortative network, training the machine with the information (i.e., time series) of low degree nodes gives better results in predicting the burst synchronization. The results are found to be consistent with the investigation carried out with a continuous time Hindmarsh-Rose neuron model. Furthermore, the role of hyperparameters like spectral radius and leaking parameter of ESN on the prediction process has been examined. Finally, we explain the underlying mechanism responsible for observing these differences in the prediction in a degree correlated network.