Predictive Approach Research Articles

Accurate Neural Network Fine-Tuning Approach for Transferable Ab Initio Energy Prediction across Varying Molecular and Crystalline Scales.

Existing machine learning models attempt to predict the energies of large molecules by training small molecules, but eventually fail to retain high accuracy as the errors increase with system size. Through an orbital pairwise decomposition of the correlation energy, a pretrained neural network model on hundred-scale data containing small molecules is demonstrated to be sufficiently transferable for accurately predicting large systems, including molecules and crystals. Our model introduces a residual connection to explicitly learn the pairwise energy corrections, and employs various low-rank retraining techniques to modestly adjust the learned network parameters. We demonstrate that with as few as only one larger molecule retraining the base model originally trained on only small molecules of (H2O)6, the MP2 correlation energy of the large liquid water (H2O)64 in a periodic supercell can be predicted at chemical accuracy. Similar performance is observed for large protonated clusters and periodic poly glycine chains. A demonstrative application is presented to predict the energy ordering of symmetrically inequivalent sublattices for distinct hydrogen orientations in the ice XV phase. Our work represents an important step forward in the quest for cost-effective, highly accurate and transferable neural network models in quantum chemistry, bridging the electronic structure patterns between small and large systems.

Predictive control approach incorporating incremental learning

Predictive control approach incorporating incremental learning

Predicting natural aging effects on fatigue life of CFRP–aluminum adhesive joints using machine learning and accelerated aging data

This study investigates the behavior and reliability of CFRP-to-aluminum adhesive joints subjected to hygrothermal conditions under both natural and accelerated aging scenarios. Natural aging durations ranged from 1 to 3 years, while accelerated aging was performed over periods of 100–1200 h in 50-hour intervals. Fatigue life was evaluated using a three-point bending test, revealing a significant degradation in joint performance due to hygrothermal exposure. Key findings include a 25.98% reduction in fatigue life for samples naturally aged for three years and a comparable 27.33% reduction in samples subjected to 1000 h of accelerated aging. Hygrothermal conditions caused notable matrix degradation, transitioning failure modes from cohesive to mixed types (cohesive, adhesive, and fiber tear failures), with a direct impact on joint durability. Machine learning models, including artificial neural network (ANN), support vector regression (SVR), linear regression, polynomial regression, and random forest regression, were employed to predict natural aging durations based on accelerated aging data. Among these, the random forest regressor exhibited the highest predictive accuracy, effectively correlating accelerated aging durations to natural aging conditions. This study provides critical insights into the failure mechanisms and long-term performance of adhesive joints, offering a novel predictive approach to estimate natural aging effects from accelerated tests. These findings highlight the potential for optimizing joint designs to enhance durability and reduce failure risks in operational environments.

On the use and misuse of time-rescaling to assess the goodness-of-fit of self-exciting temporal point processes

The paper first highlights important drawbacks and biases related to the common use of time-rescaling to assess the goodness-of-fit (Gof) of self-exciting temporal point process (SETPP) models. Then it presents a new predictive time-rescaling approach leading to an asymptotically unbiased Gof framework for general SETPPs in the case of single observed trajectories. The predictive approach focuses on forecasting accuracy and addresses the bias problem resulting from the plugged-in estimated parameters. Dawid's prequential approach is used and the models' checking is mainly based on the forecasting accuracy of arrival times. These times are transformed, using sequentially estimated parameters, into random vectors which are proved to converge in probability under the null hypothesis and standard regulatory conditions to vectors of iid Exponential(1) rv's. Numerical experiments are used to compare the performances of the standard and predictive time-rescaling for Gof assessment of non-homogeneous Poisson and Hawkes self-exciting temporal processes. Data of Japanese seismic events are also used to illustrate the dynamic aspect of the proposed model-checking approach.

The impact of temperature on the growth of Pseudomonas aeruginosa in mineral waters originated from different wells: A predictive approach

This study aimed to evaluate the behavior of Pseudomonas aeruginosa (PSA) in natural mineral water sourced from three different extraction wells and stored at various temperatures (10, 12, 20, 23, and 30 °C) to calculate the kinetic growth parameters of this microorganism through predictive modeling. The physicochemical characterization of waters was also evaluated at the time of collection, and included the analysis of 40 different minerals, and quality parameters such as pH, conductivity, oxidation-reduction potential (ORP), total dissolved solids (TDS), salinity (PSU), and temperature (T). PSA survived in raw mineral water incubated at 12, 20, 23, and 30 °C; however, no growth was observed at 10 °C. Growth curves started with an initial population of ~ 2.5–3 log CFU/mL, and final PSA populations ranged from 3.5 to 4.9 log CFU/mL. The maximum specific growth rate (μmax) at 30 °C varied among the wells, with Well P-07 showing the highest growth rate (0.2 h−1), followed by Well P-08 (0.195 h−1) and well P-01 (0.133 h−1). At 12 °C, well P-01 exhibited the highest growth rate (μmax = 0.22 h−1), indicating a influence of mineral composition in the growth of PSA. The lag time (λ) also varied, with minimum values of 2.4 ± 0.1 h at 30 °C and maximum values of 41.6 ± 0.2 h at 12 °C. From these primary estimated parameters, it was possible to obtain five robust secondary models to describe the influence of temperature on the maximum growth rates and lag phase of PSA in the well. The estimated PSA growth parameters at 20 and 23 °C were subjected to a hierarchical cluster analysis and correlation plots to verify the influence of the physicochemical composition of the waters on the PSA behavior at each well's specific annual average temperature. This analysis confirmed a positive relationship (p < 0.05) between the presence of minerals (Ca, Fe, Sr, Mn, Na) and ions (SO4−3, Cl−) and the PSA lag phase time. These results underscore the need for tailored water quality management strategies that consider chemical composition and temperature to address specific microbial contamination risks.

Artificial intelligence-based approach for water-cut prediction in crude oil processing

Artificial intelligence-based approach for water-cut prediction in crude oil processing

A Comprehensive Review on Machine Learning Approaches for Smartphone Addiction Prediction

Smartphone addiction has become a pressing public health concern, with far-reaching implications for mental health, social interactions, and overall well-being. This study proposes an unsupervised machine learning framework to predict smartphone addiction by analyzing user behavior data, such as screen time, application usage patterns, and psychological factors. The research leverages unsupervised learning techniques to identify patterns and commonalities in unlabelled data, aiming to classify individuals at risk of addiction without predefined categories. Using advanced machine learning libraries like TensorFlow and OpenCV, the system incorporates a robust pipeline involving data collection, preprocessing, and training/testing with high-performance models. The findings highlight excessive social media use and notification frequency as significant predictors, with applications ranging from personalized intervention programs and mental health monitoring to enhanced parental control systems. With a focus on scalable and adaptive methodologies, this framework not only addresses the challenges of smartphone addiction but also offers valuable insights for researchers, policymakers, and app developers to design effective solutions for healthier technology use in an increasingly digital society Key Words: Smartphone addiction, machine learning, user behavior analysis, mental health, unsupervised learning.

Experimental Verification for Machine-Learning Approaches in Compressive Strength Prediction of Alkali-Activated Concrete

Experimental Verification for Machine-Learning Approaches in Compressive Strength Prediction of Alkali-Activated Concrete

A dual deep learning approach for winter temperature prediction in solar greenhouses in Northern China

A dual deep learning approach for winter temperature prediction in solar greenhouses in Northern China

A Weighted Likelihood Ensemble Approach for Failure Prediction of Water Pipes

A Weighted Likelihood Ensemble Approach for Failure Prediction of Water Pipes

Implementing cloud computing in drug discovery and telemedicine for quantitative structure-activity relationship analysis

This work aims to use cutting-edge machine learning methods to improve quantitative structure-activity relationship (QSAR) analysis, which is used in drug development and telemedicine. The major goal is to examine the performance of several predictive modeling approaches, including random forest, deep learning-based QSAR models, and support vector machines (SVM). It investigates the potential of feature selection techniques developed in chemoinformatics for enhancing model accuracy. The innovative aspect is using cloud computing resources to strengthen computational skills, allowing for managing massive amounts of chemical information. This strategy produces accurate and generalizable QSAR models. By using the cloud's scalability and constant availability, remote healthcare apps have a workable answer. The goal is to show how these methods may improve telemedicine and the drug development process. Utilizing cloud computing equips researchers with a flexible set of tools for precise and timely QSAR analysis, speeding up the discovery of bioactive chemicals for therapeutic use. This new method fits well with the dynamic nature of pharmaceutical study and has the potential to transform the way drugs are developed and delivered to patients via telemedicine.

A physics-enhanced deep learning approach for prediction of stress intensity factors in bimaterial interface cracks

A physics-enhanced deep learning approach for prediction of stress intensity factors in bimaterial interface cracks

Physics-guided degradation trajectory modeling for remaining useful life prediction of rolling bearings

Remaining useful life (RUL) prediction has great significance in reducing operating costs and enhancing the maintainability and safety of rolling bearings. Recently, significant progress has been achieved in this field by leveraging deep learning approaches. However, advanced deep learning methods suffer from a black-box nature that makes them lack interpretability. Moreover, the prediction results may sometimes not follow laws of physics. To tackle this drawback, a physics-guided degradation trajectory modeling method is proposed for RUL prediction of rolling bearings, where physical knowledge is embedded in input preparation, model construction, and output specification. Specifically, phase space reconstruction is leveraged to construct the physics-guided inputs, transforming the degradation prediction of rolling bearings into the variation estimation of phase space trajectories. The degradation trajectory prediction strategy is derived from the physics-guided inputs and the decreased value of RUL is predicted through a compact one-dimensional convolutional neural network with the nonnegative bounded function. Simulation studies are performed with a theoretical model of defective rolling bearings to illustrate the effectiveness of the constructed physics-guided inputs. In addition, comparative analyses with advanced RUL prediction approaches under unseen working conditions and across different machines are conducted. The results demonstrate that the proposed method outperforms other approaches in both accuracy and robustness, showing its superior ability in RUL prediction of rolling bearings.

MLinvitroTox reloaded for high-throughput hazard-based prioritization of high-resolution mass spectrometry data

MLinvitroTox is an automated Python pipeline developed for high-throughput hazard-driven prioritization of toxicologically relevant signals detected in complex environmental samples through high-resolution tandem mass spectrometry (HRMS/MS). MLinvitroTox is a machine learning (ML) framework comprising 490 independent XGBoost classifiers trained on molecular fingerprints from chemical structures and target-specific endpoints from the ToxCast/Tox21 invitroDBv4.1 database. For each analyzed HRMS feature, MLinvitroTox generates a 490-bit bioactivity fingerprint used as a basis for prioritization, focusing the time-consuming molecular identification efforts on features most likely to cause adverse effects. The practical advantages of MLinvitroTox are demonstrated for groundwater HRMS data. Among the 874 features for which molecular fingerprints were derived from spectra, including 630 nontargets, 185 spectral matches, and 59 targets, around 4% of the feature/endpoint relationship pairs were predicted to be active. Cross-checking the predictions for targets and spectral matches with invitroDB data confirmed the bioactivity of 120 active and 6791 nonactive pairs while mislabeling 88 active and 56 non-active relationships. By filtering according to bioactivity probability, endpoint scores, and similarity to the training data, the number of potentially toxic features was reduced by at least one order of magnitude. This refinement makes the analytical confirmation of the toxicologically most relevant features feasible, offering significant benefits for cost-efficient chemical risk assessment.Scientific Contribution:In contrast to the classical ML-based approaches for toxicity prediction, MLinvitroTox predicts bioactivity for HRMS features (i.e., distinct m/z signals) based on MS2 fragmentation spectra rather than the chemical structures from the identified features. While the original proof of concept study was accompanied by the release of a MLinvitroTox v1 KNIME workflow, in this study, we release a Python MLinvitroTox v2 package, which, in addition to automation, expands functionality to include predicting toxicity from structures, cleaning up and generating chemical fingerprints, customizing models, and retraining on custom data. Furthermore, as a result of improvements in bioactivity data processing, realized in the concurrently released pytcpl Python package for the custom processing of invitroDBv4.1 input data used for training MLinvitroTox, the current release introduces enhancements in model accuracy, coverage of biological mechanistic targets, and overall interpretability.

Neighbourhood Component Regression Approach for Housing Unit Price Prediction

Predicting housing unit price (HUP) is important for potential buyers and investors to make informed decisions. This study proposes a novel HUP prediction model based on neighbourhood component regression (NCR). The proposed NCR model was compared with other competitive methods such as principal component regression (PCR), multiple linear regression (MLR), partial least squares regression (PLSR), and generalised linear model (GLM). When tested with real datasets, the proposed NCR method revealed prediction superiority over the four state-of-the-art methods (PCR, MLR, PLSR, and GLM). This was evident from the Mean Absolute Percentage Error (MAPE), Correlation Coefficient (R), Scatter Index (SI), and Percentage Root Mean Square Error (PRMSE) utilised as model evaluation metrics. The results revealed that the NCR model had the lowest MAPE (0.0977), SI (0.0011), PRMSE (0.1130), and highest R (0.9999) as compared with the other investigated methods. This confirms the proposed NCR method’s strength for efficient and reliable HUP prediction.

Multiclass Supervised Learning Approach for SAR-COV2 Severity and Scope Prediction: SC2SSP Framework

Multiclass Supervised Learning Approach for SAR-COV2 Severity and Scope Prediction: SC2SSP Framework

A Novel Approach for Prediction of Peak Plantar Pressure During the Gait Cycle Based on the LSTM and RNN

A Novel Approach for Prediction of Peak Plantar Pressure During the Gait Cycle Based on the LSTM and RNN

New strategy for predicting liquid–liquid equilibrium near critical point using global renormalization group theory

AbstractClassical liquid activity coefficient models, such as the nonrandom two‐liquid (NRTL) model, fail near the critical point of the liquid–liquid equilibrium (LLE), unless a highly nonlinear temperature dependency is introduced for the molecular interaction parameters. In this work, we propose an approach to predict the LLE data near the critical point using data away from the critical region based on the global renormalization group theory (GRGT). Specifically, we propose a non‐empirical approach to determine the GRGT parameters, which does not rely on experimental data. The performance of our method is examined using the NRTL model on 21 binary mixtures. Our results show that the predictive approach proposed in this work reduces the error in the critical solution temperatures by about 48% when compared to the classical NRTL model with linear temperature‐dependent interaction parameters.

A prediction technique based on deep learning for deformation and missing data in the context of insufficient sensor data

Abstract There is some interpretability in predicting deformation based on key control points, but mispredicting the deformation of the test model can result from missing sensor data due to physical obstruction and going outside the equipment’s measuring range. In order to address the issue, we propose a CNN BiLSTM CBAM network architecture. The network has a faster convergence rate, according to experiments, and its maximum deformation prediction deviation is only 0.28 mm. In the meantime, the prediction approach uses standardization to make up for the missing control point data; this results in a corrective impact that ranges from 37.76% to 90.89%. The method proposed in this paper complements the missing sensor data while predicting the deformation in real time, which is essential for establishing an accurate dynamic prediction model and realizing comprehensive data sensing.

A hybrid deep learning air pollution prediction approach based on neighborhood selection and spatio-temporal attention

Air pollution is a critical global environmental issue, further exacerbated by rapid industrialization and urbanization. Accurate prediction of air pollutant concentrations is essential for effective pollution prevention and control measures. The complex nature of pollutant data is influenced by fluctuating meteorological conditions, diverse pollution sources, and propagation processes, underscores the crucial importance of the spatial and temporal feature extraction for accurately predicting air pollutant concentrations. To address the challenges of data redundancy and diminished long-term prediction accuracy observed in previous studies, this paper presents an innovative approach to predict air pollutant concentrations leveraging advanced data analysis and deep learning methods. The proposed approach, termed KSC-ConvLSTM, integrates the k-nearest neighbors (KNN) algorithm, spatio-temporal attention (STA) mechanism, the residual block, and convolutional long short-term memory (ConvLSTM) neural network. The KNN algorithm adaptively selects highly correlated neighboring domains, while the residual block, enhanced with the STA mechanism, extracts spatial features from the input data. ConvLSTM further processes the output from STA-ConvNet to capture high-dimensional temporal and spatial features. The effectiveness of the KSC-ConvLSTM approach was validated through predictions of PM2.5 concentrations in Beijing and its surrounding urban agglomeration. The experimental results indicate that the KSC-ConvLSTM approach outperforms benchmark approaches in single-step, multi-step, and trend prediction. It demonstrates superior fitting accuracy and predictive performance. Quantitatively, the proposed KSC-ConvLSTM approach reduces the root mean square error (RMSE) by 4.216–8.458 for prediction averages of 1–12 h of PM2.5 in Beijing, compared with the benchmark approach. The findings show that the KSC-ConvLSTM approach shows considerable potential for predicting, preventing, and controlling air pollution.