Feature Importance Analysis Research Articles

Risk Identification of Mountain Torrent Hazard Using Machine Learning and Bayesian Model Averaging Techniques

Frequent mountain torrent disasters have caused significant losses to human life and wealth security and restricted the economic and social development of mountain areas. Therefore, accurate identification of mountain torrent hazards is crucial for disaster prevention and reduction. In this study, based on historical mountain torrent hazards, a mountain torrent hazard prediction model was established by using Bayesian Model Average (BMA) and three classic machine learning algorithms (gradient-boosted decision tree (GBDT), backpropagation neural network (BP), and random forest (RF)). The mountain torrent hazard condition factors used in modeling were distance to river, elevation, precipitation, slope, gross domestic product (GDP), population, and land use type. Based on the proposed BMA model, flood risk maps were produced using GIS. The results demonstrated that the BMA model significantly improved upon the accuracy and stability of single models in identifying mountain torrent hazards. The F1-values (comprehensively displays the Precision and Recall) of the BMA model under three sets of test samples at different locations were 3.31–24.61% higher than those of single models. The risk assessment results of mountain torrents found that high-risk areas were mainly concentrated in the northern border and southern valleys of Yuanyang County, China. In addition, the feature importance analysis result demonstrated that distance to river and elevation were the most important factors affecting mountain torrent hazards. The construction of projects in mountainous areas should be as far away from rivers and low-lying areas as possible. The results of this study can provide a scientific basis for improving the identification methods of mountain torrent hazards and assisting decision-makers in the implementation of appropriate measures for mountain torrent hazard prevention and reduction.

Metadata and feature importance analyses of membrane capacitive deionization models: Is a water treatment artificial intelligence panacea possible?

Membrane capacitive deionization (MCDI) is an emerging technology for water purification and desalination. Artificial intelligence (AI) techniques using machine learning (ML) and deep learning (DL) have been widely applied to overcome the drawbacks of conventional numerical models in predicting effluent concentrations, an important parameter to estimate the energy consumption and degree of desalination of the MCDI process. However, despite the remarkable progress of these ML/DL models, relevant studies have only been conducted with limited datasets, leading to a substantial controversy regarding model accuracy. Therefore, in this study, we comprehensively evaluated the performances and applicability of multiple ML/DL models for MCDI using metadata (i.e., performance indices) and feature importance analyses. One MCDI dataset containing both constant-current and constant-voltage operations was reconfigured uniformly or unevenly depending on operation modes. Subsequently, six ML models and one DL model were used to execute the analyses using performance indices. Collectively, this study suggests that a perfect ML/DL model for the water treatment process may not exist, and understanding data types is as important as selecting appropriate models to build a data-driven AI model when process performance is greatly affected by specific input variables (i.e., current and voltage).

Ensemble Learning Models for Prediction of Punching Shear Strength in RC Slab-Column Connections

In reinforced concrete (RC) structures, accurate prediction of the punching shear strength (PSS) of slab-column connections is imperative for ensuring safety. The existing equations in the literature show variability in defining parameters influencing PSS. They neglect potential variable interactions and rely on a limited dataset. This study aims to develop an accurate and reliable model to predict the PSS of slab-column connections. An extensive dataset, including 616 experimental results, was collected from earlier studies. Six robust ensemble machine learning techniques—random forest, gradient boosting, extreme gradient boosting, adaptive boosting, gradient boosting with categorical feature support, and light gradient boosting machines—are employed to predict the PSS. The findings indicate that gradient boosting stands out as the most accurate method compared to other prediction models and existing equations in the literature, achieving a coefficient of determination of 0.986. Moreover, this study utilizes techniques to explain machine learning predictions. A feature importance analysis is conducted, wherein it is observed that the reinforcement ratio and compressive strength of concrete demonstrate the highest influence on the PSS output. SHapley Additive exPlanation is conducted to represent the influence of variables on PSS. A graphical user interface for PSS prediction was developed for users’ convenience. Doi: 10.28991/CEJ-SP2024-010-01 Full Text: PDF

Optimization and prediction of catalysts for precise synthesis of methyl glycolate from dimethyl oxalate using machine learning coupled with particle swarm optimization algorithm

To optimize and predict the catalyst for precise synthesis of methyl glycolate (MG) from dimethyl oxalate, this study proposed a six-step machine learning framework coupled with a particle swarm optimization algorithm. The random forest (RF) model has the highest prediction accuracy after optimizing its hyperparameters. This preferred model has been rigorously validated using experimental data, showcasing a remarkable level of consistency with the observed trends. Then the catalytic performance of the dimethyl oxalate to methyl glycolate (DtMG) process is evaluated by the feature importance analysis and partial dependence plot approaches. It is recommended to operate at lower temperatures and pressures, higher space velocities, and hydrogen-to-ester ratio for MG production. Finally, the RF model coupled with the particle swarm optimization algorithm is employed to predict the optimal catalyst for maximizing MG yield and minimizing cost of the DtMG process, which successfully predicts seven new catalysts with higher yields and lower costs.

Development and validation of a machine learning-based prediction model for sudden cardiac death in the general population: insights from the community cardiovascular cohort

Abstract Background/Introduction Accurately predicting and managing the risk of sudden cardiac death (SCD) in the general population is a formidable challenge. Purpose This study aimed to develop and validate a machine learning (ML)-based prediction model for SCD using data from the Community Cardiovascular Cohort (CCCC) in Taiwan. Methods A community-based cohort study was used for model construction . Four ML algorithms, including extreme gradient boosting (XGBoost), random forests (RF), logistic regression, and deep neural network (DNN), underwent meticulous evaluation. Performance assessment encompassed receiver operating characteristic (ROC) curves, area under the curves (AUC), and accuracy metrics. Significant risk factors were identified through feature importance analysis. Results We employed a random partitioning strategy, allocating 1744 participants (55%) to the training dataset, 952 participants (30%) to the test dataset, and 476 participants (15%) to the validation dataset. In the data training stage, the XGBoost algorithm demonstrated outstanding prediction performance for SCD (AUC: 1.000); In the validation stage, the XGBoost algorithm demonstrated good prediction performance for SCD prediction (AUC: 0.791, 95%CI: 0.617-0.964; Accuracy: 0.971). Using XGBoost-generated predicted probabilities (P(XGB)), we identified individuals with P(XGB) ≥ 0.5 (Group 1) who displayed significantly higher SCD risk (HR 32.0; 95% CI: 9.29-109.9) compared to those with P(XGB) < 0.5 (Group 0). Conclusions Our study demonstrates the effectiveness of ML algorithms, particularly XGBoost, in predicting event of SCD in the general population. The identification of these pivotal risk factors holds profound implications, empowering the identification of high-risk individuals and facilitating the implementation of precisely tailored preventive strategies.Figure 1:KM sruvival curveFigure 2:Feature importance analysis

Application of Machine Learning to Research on Trace Elemental Characteristics of Metal Sulfides in Se-Te Bearing Deposits

This study focuses on exploring the indication and importance of selenium (Se) and tellurium (Te) in distinguishing different genetic types of ore deposits. Traditional views suggest that dispersed elements are unable to form independent deposits, but are hosted within deposits of other elements as associated elements. Based on this, the study collected trace elemental data of pyrite, sphalerite, and chalcopyrite in various types of Se-Te bearing deposits. The optimal end-elements for distinguishing different genetic type deposits were recognized by principal component analysis (PCA) and the silhouette coefficient method, and discriminant diagrams were drawn. However, support vector machine (SVM) calculation of the decision boundary shows low accuracy, revealing the limitations in binary discriminant visualization for ore deposit type discrimination. Consequently, two machine learning algorithms, random forest (RF) and SVM, were used to construct ore genetic type classification models on the basis of trace elemental data for the three types of metal sulfides. The results indicate that the RF classification model for pyrite exhibits the best performance, achieving an accuracy of 94.5% and avoiding overfitting errors. In detail, according to the feature importance analysis, Se exhibits higher Shapley Additive Explanations (SHAP) values in volcanogenic massive sulfide (VMS) and epithermal deposits, especially the latter, where Se is the most crucial distinguishing element. By comparison, Te shows a significant contribution to distinguishing Carlin-type deposits. Conversely, in porphyry- and skarn-type deposits, the contributions of Se and Te were relatively lower. In conclusion, the application of machine learning methods provides a novel approach for ore genetic type classification and discrimination research, enabling more accurate identification of ore genetic types and contributing to the exploration and development of mineral resources.

Towards greener futures: SVR-based CO2 prediction model boosted by SCMSSA algorithm

This research presents the utilization of an enhanced Sine cosine perturbation with Chaotic perturbation and Mirror imaging strategy-based Salp Swarm Algorithm (SCMSSA), which incorporates three improvement mechanisms, to enhance the convergence accuracy and speed of the optimization algorithm. The study assesses the SCMSSA algorithm's performance against other optimization algorithms using six test functions to show the efficacy of the enhancement strategies. Furthermore, its efficacy in improving Support Vector Regression (SVR) models for CO2 prediction is assessed. The results reveal that the SVR-SCMSSA hybrid model surpasses other hybrid models and standard SVR in terms of training and prediction accuracy by obtaining 95 % accuracy. Its swift convergence, precision, and resistance to local optima position make it an excellent choice for addressing complex problems such as CO2 prediction, with critical implications for sustainability efforts. Moreover, feature importance analysis by SVR-SCMSSA offers valuable insights into the key contributors to CO2 prediction in the dataset, emphasizing the significance and impact of factors such as fossil fuel, Biomass, and Wood as major contributors to CO2 emission. The research suggests the adoption of the SVR-SCMSSA hybrid model for more accurate and reliable CO2 prediction to researchers and policymakers, which is essential for environmental sustainability and climate change mitigation.

Active Learning of Ligands That Enhance Perovskite Nanocrystal Luminescence.

Ligands play a critical role in the optical properties and chemical stability of colloidal nanocrystals (NCs), but identifying ligands that can enhance NC properties is daunting, given the high dimensionality of chemical space. Here, we use machine learning (ML) and robotic screening to accelerate the discovery of ligands that enhance the photoluminescence quantum yield (PLQY) of CsPbBr3 perovskite NCs. We developed a ML model designed to predict the relative PL enhancement of perovskite NCs when coordinated with a ligand selected from a pool of 29,904 candidate molecules. Ligand candidates were selected using an active learning (AL) approach that accounted for uncertainty quantified by twin regressors. After eight experimental iterations of batch AL (corresponding to 21 initial and 72 model-recommended ligands), the uncertainty of the model decreased, demonstrating an increased confidence in the model predictions. Feature importance and counterfactual analyses of model predictions illustrate the potential use of ligand field strength in designing PL-enhancing ligands. Our versatile AL framework can be readily adapted to screen the effect of ligands on a wide range of colloidal nanomaterials.

Deep learning-based prediction of structural responses of RC slabs subjected to blast loading

Considering the risk of explosion, studying the dynamic structural behavior of reinforced concrete (RC) slabs subjected to blast loads is crucial for enhancing their resistance. Conventional approaches, including costly experimental tests, highly hypothetical analytical methods, and time-consuming numerical simulations, have their limitations. This study presents two deep learning-based models for rapid and accurate prediction of explosion-induced responses in RC slabs. Available literature data and supplemented numerical simulation data are used for training and testing. First, a multi-layer perceptron (MLP) model is established to predict the maximum displacement of RC slabs subjected to explosions. The input features include 10 parameters. The training results show that the MLP model exhibits good prediction performance. Second, a one-dimensional convolutional neural network (1D-CNN) model is constructed to predict the failure modes of RC slabs under explosive loads. The input features of the model are consistent with those of the MLP model. The 1D-CNN model outperforms the other five conventional machine learning models in classification. Furthermore, a permutation feature importance analysis is conducted to determine the effects of the 10 input features on the predicted outcomes. This analysis makes the predictions interpretable, thereby bolstering the credibility of the models. Results show that the proposed deep learning models offer an efficient and reliable method to predict the structural response of RC slabs under explosion loads.

Machine learning-based model to predict delirium in patients with advanced cancer treated with palliative care: a multicenter, patient-based registry cohort

This study aimed to present a new approach to predict to delirium admitted to the acute palliative care unit. To achieve this, this study employed machine learning model to predict delirium in patients in palliative care and identified the significant features that influenced the model. A multicenter, patient-based registry cohort study in South Korea between January 1, 2019, and December 31, 2020. Delirium was identified by reviewing the medical records based on the criteria of the Diagnostic and Statistical Manual of Mental Disorders, Fifth Edition. The study dataset included 165 patients with delirium among 2314 patients with advanced cancer admitted to the acute palliative care unit. Seven machine learning models, including extreme gradient boosting, adaptive boosting, gradient boosting, light gradient boosting, logistic regression, support vector machine, and random forest, were evaluated to predict delirium in patients with advanced cancer admitted to the acute palliative care unit. An ensemble approach was adopted to determine the optimal model. For k-fold cross-validation, the combination of extreme gradient boosting and random forest provided the best performance, achieving the following accuracy metrics: 68.83% sensitivity, 70.85% specificity, 69.84% balanced accuracy, and 74.55% area under the receiver operating characteristic curve. The performance of the isolated testing dataset was also validated, and the machine learning model was successfully deployed on a public website (http://ai-wm.khu.ac.kr/Delirium/) to provide public access to delirium prediction results in patients with advanced cancer. Furthermore, using feature importance analysis, sex was determined to be the top contributor in predicting delirium, followed by a history of delirium, chemotherapy, smoking status, alcohol consumption, and living with family. Based on a large-scale, multicenter, patient-based registry cohort, a machine learning prediction model for delirium in patients with advanced cancer was developed in South Korea. We believe that this model will assist healthcare providers in treating patients with delirium and advanced cancer.

Mechanism of targeted adsorption and specific action of bone char on ionic organic pollutants: An analysis based on differences in physicochemical properties

Adsorption is a widely recognized and effective method for treating high-concentration organic industrial wastewater. Bone char (BC) is a promising adsorbent for removing ionic organic pollutants (IOPs) from printing and dyeing wastewater due to its advantageous adsorption properties. To analyze BC's targeted adsorption and specific action mechanism on IOPs, the author first elucidated the physicochemical properties of BC under different preparation conditions. Subsequently, the study compared the targeted adsorption efficacy of BC on various types and molecular structures of IOPs. Through correlation analysis, feature importance analysis, and consideration of the properties of the target substance, the study investigated the targeted adsorption laws of BC for IOPs. Lastly, for substances with specific adsorption characteristics, range analysis, structure–activity relationship analysis, and quantitative analysis of adsorption mechanisms were employed to elucidate the action mechanism of BC. The results indicate that the differences in physicochemical properties (especially CEC and pH) caused by the inorganic mineral components of BC and the number of conjugated double bonds in the molecular structure of IOPs are key factors determining the adsorption effect. Controlling BC's pore structure and electrical conductivity based on the aggregates' conjugated double bonds and particle size can improve adsorption outcomes. The hierarchy of preparation factors influencing the adsorption performance of BC is pyrolysis temperature > particle size > pyrolysis atmosphere. The adsorption of malachite green (MG) by BC primarily involves chemical reactions, including mineral complexation effect (52.38 %), electrostatic interaction (24.76 %), π-π interaction (8.15 %), metal ion exchange (8.12 %), and physical pore filling (6.59 %). It is speculated that the primary interaction mechanism between BC and MG is through cation-π bonds formed between metal cations dissociated from the inorganic mineral components of BC and MG, facilitated by strong electrostatic interaction. The findings offer a theoretical basis for the practical utilization of BC in adsorbing IOPs.

Determinants of Yearly CO2 Emission Fluctuations: A Machine Learning Perspective to Unveil Dynamics

In order to understand the dynamics in climate change, inform policy decisions and prompt timely action to mitigate its impact, this study provides a comprehensive analysis of the short-term trend of the year-on-year CO2 emission changes across ten countries, considering a broad range of factors including socioeconomic factors, CO2-related industry, and education. This study uniquely goes beyond the common country-based analysis, offering a broader understanding of the interconnected impact of CO2 emissions across countries. Our preliminary regression analysis, using the ten most significant features, could only explain 66% of the variations in the target. To capture the emissions trend variation, we categorized countries by the change in CO2 emission volatility (high, moderate, low with upward or downward trends), assessed using standard deviation. We employed machine learning techniques, including feature importance analysis, Partial Dependence Plots (PDPs), sensitivity analysis, and Pearson and Canonical correlation analyses, to identify influential factors driving these short-term changes. The Decision Tree Classifier was the most accurate model, with an accuracy of 96%. It revealed population size, CO2 emissions from coal, the three-year average change in CO2 emissions, GDP, CO2 emissions from oil, education level (incomplete primary), and contribution to temperature rise as the most significant predictors, in order of importance. Furthermore, this study estimates the likelihood of a country transitioning to a higher emission category. Our findings provide valuable insights into the temporal dynamics of factors influencing CO2 emissions changes, contributing to the global efforts to address climate change.

Combining Different Wearable Devices to Assess Gait Speed in Real-World Settings.

Assessing mobility in daily life can provide significant insights into several clinical conditions, such as Chronic Obstructive Pulmonary Disease (COPD). In this paper, we present a comprehensive analysis of wearable devices' performance in gait speed estimation and explore optimal device combinations for everyday use. Using data collected from smartphones, smartwatches, and smart shoes, we evaluated the individual capabilities of each device and explored their synergistic effects when combined, thereby accommodating the preferences and possibilities of individuals for wearing different types of devices. Our study involved 20 healthy subjects performing a modified Six-Minute Walking Test (6MWT) under various conditions. The results revealed only little performance differences among devices, with the combination of smartwatches and smart shoes exhibiting superior estimation accuracy. Particularly, smartwatches captured additional health-related information and demonstrated enhanced accuracy when paired with other devices. Surprisingly, wearing all devices concurrently did not yield optimal results, suggesting a potential redundancy in feature extraction. Feature importance analysis highlighted key variables contributing to gait speed estimation, providing valuable insights for model refinement.

Predictive Analysis of Campus Placement of Student Using Machine Learning Algorithms

In today's competitive job market, student campus placement process becomes very important role in the academic journey of students and the reputation of educational institutions alike. As the demand for skilled professionals continues to rise, the ability to accurately predict and optimize student placement outcomes becomes increasingly essential. This research paper proposes a machine learning algorithms analysis to forecast the campus placement of students with the help of various academic factors. The dataset contains historical placement records of students received from open-source repository named as kaggle. It includes information such as student academic achievements, internship, project, workshop, soft skill, extracurricular activities, and placement outcomes. Through data preprocessing and feature engineering, we transform raw data into a structured format suitable for predictive modeling. One hot encoding (Dummy encoding) applied on dataset to represent categorical variables as numerical values to provide more information to the model about the categorical variables. Predictive analysis work on various machine learning algorithms, including logistic regression, decision trees, random forests, support vector machines (SVM), XGBoost, AdaBoost, KNN and Naive Byes. The results of analysis demonstrate in predicting student placement outcomes with high accuracy and reliability. Accuracy is calculated in the form of precision, recall, f1-score, and support. Furthermore, we conduct feature importance analysis to identify the importance of different predictors in determining placement success. This research paper study can help to academic institutions that can identify at-risk students early in their academic journey and to improve their employability. In conclusion, research underscores the potential of predictive analysis and machine learning in optimizing student placement outcomes. By harnessing the power of data-driven insights, institutes can make use of placement prediction model to check the likelihood of campus placement of students.

Decision tree-based approach to extrapolate life cycle inventory data of manufacturing processes

Life cycle assessment (LCA) plays a crucial role in green manufacturing to uncover the critical aspects for alleviating the environmental burdens due to manufacturing processes. However, the scarcity of life cycle inventory (LCI) data for the manufacturing processes is a considerable challenge. This paper proposes a novel approach to extrapolate LCI data of manufacturing processes. Taking advantage of LCI data in the Ecoinvent datasets, decision tree-based supervised machine learning models, namely decision tree, random forest, gradient boosting, and adaptive boosting, have been developed to extrapolate the data of GHG emissions, i.e., carbon dioxide, nitrous oxide, methane, and water vapor. Initially, a correlation analysis was conducted to derive the most influential factors on GHG quantities resulting from manufacturing activities. First, the collected data have been preprocessed and split into train and test sets (70% and 30%, respectively). Second, a five-fold cross-validation method was applied to tune the hyperparameters of the models. Then, the models were re-trained using the best hyperparameters and evaluated using the test set. The results reveal that the Gradient Boosting model has a superior predictive performance for extrapolating the GHG emission data, with average coefficients of determination (R2) on the test set <0.95. Moreover, the model predictions involve relatively low values of the average root mean squared error and an average mean percentage of error on the test set. The correlation and feature importance analyses emphasized that the workpiece material and manufacturing technology have a considerable effect on natural resource consumption, i.e., energy, material, and water inflows into the process. Meanwhile, energy consumption, water usage, and raw aluminum depletion were the most influential factors in GHG emissions. Eventually, a case study to extrapolate the inflows and the outflows for new manufacturing activities has been conducted using the validated models. The proposed GraBoost model provides a computational supplementary approach to estimate and extrapolate the GHG emissions for different manufacturing processes when LCI data are incomplete or don't exist within LCI databases.

ThermoFinder: A sequence-based thermophilic proteins prediction framework

Thermophilic proteins are important for academic research and industrial processes, and various computational methods have been developed to identify and screen them. However, their performance has been limited due to the lack of high-quality labeled data and efficient models for representing protein. Here, we proposed a novel sequence-based thermophilic proteins prediction framework, called ThermoFinder. The results demonstrated that ThermoFinder outperforms previous state-of-the-art tools on two benchmark datasets, and feature ablation experiments confirmed the effectiveness of our approach. Additionally, ThermoFinder exhibited exceptional performance and consistency across two newly constructed datasets, one of these was specifically constructed for the regression-based prediction of temperature optimum values directly derived from protein sequences. The feature importance analysis, using shapley additive explanations, further validated the advantages of ThermoFinder. We believe that ThermoFinder will be a valuable and comprehensive framework for predicting thermophilic proteins, and we have made our model open source and available on Github at https://github.com/Luo-SynBioLab/ThermoFinder.

Evaluating spray drift from Uncrewed Aerial Spray Systems: A machine learning and variance-based sensitivity analysis of environmental and spray system parameters

Uncrewed Aerial Spray Systems (UASS), commonly called drones, have become an important application technique for plant protection products in Asia and worldwide. As such, environmental variables and spray system parameters influencing spray drift deserve detailed investigations. This study presents the data analysis of 114 UASS drift trials conducted between December 2021 and December 2022 in China. Study design was based on the ISO 22866:2005 protocol for spray drift trials and considered different UASS platforms, nozzles, and release heights, and specifically continuously measured weather conditions. The relative importance of the environmental variables and spray system parameters was evaluated by a random forest (RF) feature importance analysis, a Sobol sensitivity analysis and partial dependence plots. This approach was preferred to linear ranking techniques such as ANOVA (analysis of variance) due to the non-linearity of the system. In addition, partial dependence plots are proposed to visualize the relationship between specific input parameters within the system.Drift deposition curves calculated from the 114 trials show good agreement with previous UASS trials reported in the literature. As reported in previous studies, spray drift following UASS applications is lower than for manned aerial vehicles, greater than for ground spray applications, and similar to drift observed from orchard air blast applications. In addition, 9 trials were conducted on corn fields in order to evaluate the potential effect of crop cover on spray drift. Spray drift was observed to be reduced over the cropped soil, suggesting that plant cover might possibly reduce spray drift. These findings could help supporting drift mitigation policies, stewardship advice and product labelling around the world.

PRECYSE: Predicting Cybersickness using Transformer for Multimodal Time-Series Sensor Data

Cybersickness, a factor that hinders user immersion in VR, has been the subject of ongoing attempts to predict it using AI. Previous studies have used CNN and LSTM for prediction models and used attention mechanisms and XAI for data analysis, yet none explored a transformer that can better reflect the spatial and temporal characteristics of the data, beneficial for enhancing prediction and feature importance analysis. In this paper, we propose cybersickness prediction models using multimodal time-series sensor data (i.e., eye movement, head movement, and physiological signals) based on a transformer algorithm, considering sensor data pre-processing and multimodal data fusion methods. We constructed the MSCVR dataset consisting of normalized sensor data, spectrogram formatted sensor data, and cybersickness levels collected from 45 participants through a user study. We proposed two methods for embedding multimodal time-series sensor data into the transformer: modality-specific spatial and temporal transformer encoders for normalized sensor data (MS-STTN) and modality-specific spatial-temporal transformer encoder for spectrogram (MS-STTS). MS-STTN yielded the highest performance in the ablation study and the comparison of the existing models. Furthermore, by analyzing the importance of data features, we determined their relevance to cybersickness over time, especially the salience of eye movement features. Our results and insights derived from multimodal time-series sensor data and the transformer model provide a comprehensive understanding of cybersickness and its association with sensor data. Our MSCVR dataset and code are publicly available: https://github.com/dayoung-jeong/PRECYSE.git.

Enhanced Bearing Fault Diagnosis Through Trees Ensemble Method and Feature Importance Analysis

Abstract Purpose This research introduces a groundbreaking method for bearing defect detection. It leverages ensemble machine learning (ML) models and conducts comprehensive feature importance analysis. The key innovation is the training and benchmarking of three tree ensemble models—Decision Tree (DT), Random Forest (RF), and Extreme Gradient Boosting (XGBoost)—on an extensive experimental dataset (QU-DMBF) collected from bearing tests with seeded defects of varying sizes on the inner and outer raceways under different operating conditions. Method The dataset was meticulously prepared with categorical variable encoding and Min–Max data normalization to ensure consistent class distribution and model accuracy. Implementing the ML models involved a grid search method for hyperparameter tuning, focusing on reporting the models’ accuracy. The study also explores applying ensemble methods and using supervised and unsupervised learning algorithms for bearing fault detection. It underscores the value of feature importance analysis in understanding the contributions of specific inputs to the model’s performance. The research compares the ML models to traditional methods and discusses their potential for advanced fault diagnosis in bearing systems. Results and Conclusions The XGBoost model, trained on data from actual bearing tests, outperformed the others, achieving 92% accuracy in detecting bearing health and fault location. However, a deeper analysis of feature importance reveals that the models weigh certain experimental conditions differently—such as sensor location and motor speed. This research’s primary novelties and contributions are comparative evaluation, experimental validation, accuracy benchmarking, and interpretable feature importance analysis. This comprehensive methodology advances the bearing health monitoring field and has significant practical implications for condition-based maintenance, potentially leading to substantial cost savings and improved operational efficiency.

MoodCapture: Depression Detection Using In-the-Wild Smartphone Images.

MoodCapture presents a novel approach that assesses depression based on images automatically captured from the front-facing camera of smartphones as people go about their daily lives. We collect over 125,000 photos in the wild from N=177 participants diagnosed with major depressive disorder for 90 days. Images are captured naturalistically while participants respond to the PHQ-8 depression survey question: "I have felt down, depressed, or hopeless". Our analysis explores important image attributes, such as angle, dominant colors, location, objects, and lighting. We show that a random forest trained with face landmarks can classify samples as depressed or non-depressed and predict raw PHQ-8 scores effectively. Our post-hoc analysis provides several insights through an ablation study, feature importance analysis, and bias assessment. Importantly, we evaluate user concerns about using MoodCapture to detect depression based on sharing photos, providing critical insights into privacy concerns that inform the future design of in-the-wild image-based mental health assessment tools.