ML Models Research Articles

Machine-learning-based prediction by stacking ensemble strategy for surgical outcomes in patients with degenerative cervical myelopathy

BackgroundMachine learning (ML) is extensively employed for forecasting the outcome of various illnesses. The objective of the study was to develop ML based classifiers using a stacking ensemble strategy to predict the Japanese Orthopedic Association (JOA) recovery rate for patients with degenerative cervical myelopathy (DCM).MethodsA total of 672 patients with DCM were included in the study and labeled with JOA recovery rate by 1-year follow-up. All data were collected during 2012–2023 and were randomly divided into training and testing (8:2) sub-datasets. A total of 91 initial ML classifiers were developed, and the top 3 initial classifiers with the best performance were further stacked into an ensemble classifier with a supported vector machine (SVM) classifier. The area under the curve (AUC) was the main indicator to assess the prediction performance of all classifiers. The primary predicted outcome was the JOA recovery rate.ResultsBy applying an ensemble learning strategy (e.g., stacking), the accuracy of the ML classifier improved following combining three widely used ML models (e.g., RFE-SVM, EmbeddingLR-LR, and RFE-AdaBoost). Decision curve analysis showed the merits of the ensemble classifiers, as the curves of the top 3 initial classifiers varied a lot in predicting JOA recovery rate in DCM patients.ConclusionsThe ensemble classifiers successfully predict the JOA recovery rate in DCM patients, which showed a high potential for assisting physicians in managing DCM patients and making full use of medical resources.

Lateral cephalometric parameters among Arab skeletal classes II and III patients and applying machine learning models

BackgroundThe World Health Organization considers malocclusion one of the most essential oral health problems. This disease influences various aspects of patients’ health and well-being. Therefore, making it easier and more accurate to understand and diagnose patients with skeletal malocclusions is necessary.ObjectivesThe main aim of this research was the establishment of machine learning models to correctly classify individual Arab patients, being citizens of Israel, as skeletal class II or III. Secondary outcomes of the study included comparing cephalometric parameters between patients with skeletal class II and III and between age and gender-specific subgroups, an analysis of the correlation of various cephalometric variables, and principal component analysis in skeletal class diagnosis.MethodsThis quantitative, observational study is based on data from the Orthodontic Center, Jatt, Israel. The experimental data consisted of the coded records of 502 Arab patients diagnosed as Class II or III according to the Calculated_ANB. This parameter was defined as the difference between the measured ANB angle and the individualized ANB of Panagiotidis and Witt. In this observational study, we focused on the primary aim, i.e., the establishment of machine learning models for the correct classification of skeletal class II and III in a group of Arab orthodontic patients. For this purpose, various ML models and input data was tested after identifying the most relevant parameters by conducting a principal component analysis. As secondary outcomes this study compared the cephalometric parameters and analyzed their correlations between skeletal class II and III as well as between gender and age specific subgroups.ResultsComparison of the two groups demonstrated significant differences between skeletal class II and class III patients. This was shown for the parameters NL-NSL angle, PFH/AFH ratio, SNA angle, SNB angle, SN-Ba angle. SN-Pg angle, and ML-NSL angle in skeletal class III patients, and for S-N (mm) in skeletal class II patients. In skeletal class II and skeletal class III patients, the results showed that the Calculated_ANB correlated well with many other cephalometric parameters. With the help of the Principal Component Analysis (PCA), it was possible to explain about 71% of the variation between the first two PCs. Finally, applying the stepwise forward Machine Learning models, it could be demonstrated that the model works only with the parameters Wits appraisal and SNB angle was able to predict the allocation of patients to either skeletal class II or III with an accuracy of 0.95, compared to a value of 0.99 when all parameters were used (“general model”).ConclusionThere is a significant relationship between many cephalometric parameters within the different groups of gender and age. This study highlights the high accuracy and power of Wits appraisal and the SNB angle in evaluating the classification of orthodontic malocclusion.

Comparison of a machine learning model with a conventional rule-based selective dry cow therapy algorithm for detection of intramammary infections

We trained machine learning models to identify intramammary infections (IMI) in late lactation cows at dry-off to guide antibiotic treatment, and compared their performance to a rule-based algorithm that is currently used on dairy farms in the US. We conducted an observational test-characteristics study using a data set of 3,645 cows approaching dry-off from 68 US dairy herds. The outcome variables of interest were cow-level IMI caused by all pathogens, major pathogens, and Streptococcus and Strep-like organisms (SSLO), which were determined using aerobic culture of aseptic quarter-milk samples and identification of isolates using MALDI-TOF. Individual cow records were extracted from the farm software to create 53 feature variables at the cow and 39 at the herd-level which were derived from cow-level descriptive data, records of clinical mastitis events, results from routine testing of milk for volume and concentrations of somatic cell count (SCC), fat, and protein. ML algorithms evaluated were logistic regression, decision tree, random forest, light gradient-boosting machine, naïve bayes, and neural networks. For comparison, cows were also classified according to a conventional rule-based algorithm that considered a cow as high risk for IMI if she had at one or more high SCC (>200,000 cells/ml) tests or ≥2 cases of clinical mastitis during the lactation of enrollment. Area under the curve (AUC) and Youden's index were used to compare models, in addition to binary classification metrics, including sensitivity, specificity, and predictive values. ML models had slightly higher AUC and Youden's index values than the rule-based algorithm for all IMI outcomes of interest. However, these improvements in prediction accuracy were substantially less than what we had considered necessary for the technology to be a worthwhile alternative to the rule-based algorithm. Therefore, evidence is lacking to support the wholesale use of ML-guided selective dry cow therapy at the moment. We recommend that producers wanting to implement algorithm-guided SDCT use a rule-based method.

Robustness Investigation of ML Models for Stock Price Prediction During Market Volatility: A Case Study of the Japanese Stock Market Amidst the COVID-19 Pandemic

In the evolving realm of financial markets, precise stock price prediction remains a pivotal challenge. This paper investigates the efficacy of traditional machine learning models for stock price forecasting, particularly during the high volatility phase induced by the COVID-19 pandemic, using the Japanese stock market as a case study. This study utilized a dataset from the JPX Tokyo Stock Exchange Prediction competition, selecting two stocks at random to avoid bias. After standardizing the data for consistency, this study employed linear regression and decision trees to project stock prices, comparing their performance during stable periods and the economic upheaval of the pandemic. The analysis revealed that both models' predictive capabilities were compromised during the pandemic, with the decision tree model particularly underperforming. In normal times, linear regression showed improved accuracy for stock 1301 in 2021, while for stock 6758, the model's performance declined. These findings underscore the necessity for robust models that can withstand market turbulences like those observed during the COVID-19 crisis. The research contributes to both academic discussions on financial forecasting and practical strategies for market participants navigating complex stock environments. Future work will focus on developing more resilient forecasting methods to address these challenges.

Flood risk decomposed: Optimized machine learning hazard mapping and multi-criteria vulnerability analysis in the city of Zaio, Morocco

Urban flood risk mapping has become crucial for effective mitigation and urban planning. This study assesses and maps flood risk in the city of Zaio, Morocco, using machine learning and Multi-Criteria Decision Analysis (MCDA) techniques to overcome data scarcity challenges. We employed the Random Forest (RF) model with nine flood conditioning factors for flood hazard and the Analytical Hierarchy Process (AHP) for vulnerability assessment. To enhance RF model performance, we compared three hyperparameter tuning techniques: Bayesian Optimization (RF-BO), Genetic Algorithm (RF-GA), and Grid Search (RF-GS). RF-BO demonstrated superior accuracy in flood hazard modelling. Flood vulnerability was assessed using AHP, incorporating social and physical factors. The final flood risk map was produced by combining the RF-BO hazard model with the AHP vulnerability assessment. Results indicate that flood hazard in Zaio is significantly affected by factors such as topography and distance to rivers. Moreover, areas with high population density closer to rivers, especially in the south-western residential area, were found to be more exposed to flood risk. The findings highlight the utility of ML models, MCDA, and hyperparameter optimization in urban flood risk mapping, enabling the identification of high-risk urban areas that should be prioritized for flood protection efforts. This approach proves especially valuable in ungauged regions with limited hydrological data.

Surgical control time estimation variability: Implications for medical systems and the future integration of AI and ML models

BackgroundAccurate estimation of surgical procedure times, crucial for optimizing healthcare access, patient outcomes, and cost-effectiveness, is essential for operating room efficiency. Surgical control time (SCT) is a preoperative estimate by surgeons representing their predicted time to complete the surgery, spanning from completion of anesthesia induction to surgical site closure. MethodsIn this within-subjects, longitudinal study, we examined the differences between predicted surgical control times versus actual SCTs and determined variability by surgical specialty. We included cases regardless of classification (i.e., outpatient or inpatient), type of surgery (i.e., elective, urgent, or emergent), or level of complexity (i.e., major or minor). We ran Shapiro–Wilk tests to assess the normality of the difference in actual versus predicted surgical control times (dSCT) by surgical specialty. We used a generalized linear model (GLM) with robust clustered variance and pairwise comparisons of surgical specialties (with Bonferroni adjustment for family-wise error rate) to assess differences in the prediction accuracy of SCTs by specialty. ResultsWe analyzed 14,438 surgical cases performed by 168 surgeons across 13 specialties from January 2019 to January 2023. 11 of 13 specialties had higher actual than predicted times, suggesting an overall pattern of underestimating SCTs. On average, surgeries took 12.3 % longer than predicted, with surgeons underestimating SCTs by an average of 10.4 min. SCTs comprised 78 % of the total operative time. The four specialties with the largest underestimations of SCTs were neurosurgery (27.04 min), orthopedics (22.75 min), urology (19.4 min, and plastic surgery (18.67 min), while two specialties exhibited overestimations, namely ear nose and throat (11.14 min) and pediatrics (–3.21 min). GLM results and pairwise comparisons showed that surgeons significantly differed in their SCT prediction by surgical specialty. ConclusionsOur findings showed significant differences across surgical specialties in the accuracy of predicting surgical control times. These results have implications for integrating evolving technologies such as artificial intelligence and machine learning models to assist surgical administrators in accurately predicting surgical case durations and optimizing resource allocation.

Interpretable prediction of acute ischemic stroke after hip fracture in patients 65 years and older based on machine learning and SHAP

BackgroundHip fracture and acute ischemic stroke (AIS) are prevalent conditions among the older population. The prognosis for older patients who experience AIS subsequent to hip fracture is frequently unfavorable. MethodsPatients were categorized into the AIS group and the non-AIS group. A predictive model was developed using six different machine learning algorithms. The SHapley Additive exPlanations (SHAP) method was then utilized to provide both local and global explanations. We performed adjusted mediation analyses. Furthermore, a nomogram was created to present the outcomes obtained from the LASSO regression examination. The main objective was to ascertain influential elements that can predict the occurrence of AIS. To alleviate the influence of confounding variables, propensity score matching was utilized to compare the occurrence of additional complications. Survival was compared by Kaplan-Meier methods. ResultsThe AUC of 6 ML models ranged from 0.73 to 0.87. The SVM model exhibited the greatest efficacy in forecasting AIS among older individuals with hip fractures. The leading 6 variables in the support vector machines (SVM) model were identified as systemic inflammatory response index (SIRI), carotid atherosclerosis, prior stroke, C-reactive protein (CRP), fibrinogen (FIB), and hypertension. The leading 2 variables in SHAP were identified as FIB at admission and SIRI index. There wasn't potential mediating effect of admission FIB between the SIRI index and AIS. There were statistically significant differences between the two groups in survival (P=0.003). ConclusionsThe model displayed good performance for prediction of AIS after hip fracture in patients 65 years and older, which might facilitate to establishment of a better clinical assessment plan.

Identification of putative coral pathogens in endangered Caribbean staghorn coral using machine learning.

Coral diseases contribute to the rapid decline in coral reefs worldwide, and yet coral bacterial pathogens have proved difficult to identify because 16S rRNA gene surveys typically identify tens to hundreds of disease-associate bacteria as putative pathogens. An example is white band disease (WBD), which has killed up to 95% of the now-endangered Caribbean Acropora corals since 1979, yet the pathogen is still unknown. The 16S rRNA gene surveys have identified hundreds of WBD-associated bacterial amplicon sequencing variants (ASVs) from at least nine bacterial families with little consensus across studies. We conducted a multi-year, multi-site 16S rRNA gene sequencing comparison of 269 healthy and 143 WBD-infected Acropora cervicornis and used machine learning modelling to accurately predict disease outcomes and identify the top ASVs contributing to disease. Our ensemble ML models accurately predicted disease with greater than 97% accuracy and identified 19 disease-associated ASVs and five healthy-associated ASVs that were consistently differentially abundant across sampling periods. Using a tank-based transmission experiment, we tested whether the 19 disease-associated ASVs met the assumption of a pathogen and identified two pathogenic candidate ASVs-ASV25 Cysteiniphilum litorale and ASV8 Vibrio sp. to target for future isolation, cultivation, and confirmation of Henle-Koch's postulate via transmission assays.

Eliciting Model Steering Interactions From Users via Data and Visual Design Probes.

Visual and interactive machine learning systems (IML) are becoming ubiquitous as they empower individuals with varied machine learning expertise to analyze data. However, it remains complex to align interactions with visual marks to a user's intent for steering machine learning models. We explore using data and visual design probes to elicit users' desired interactions to steer ML models via visual encodings within IML interfaces. We conducted an elicitation study with 20 data analysts with varying expertise in ML. We summarize our findings as pairs of target-interaction, which we compare to prior systems to assess the utility of the probes. We additionally surfaced insights about factors influencing how and why participants chose to interact with visual encodings, including refraining from interacting. Finally, we reflect on the value of gathering such formative empirical evidence via data and visual design probes ahead of developing IML prototypes.

Secure cognitive health monitoring using a directed acyclic graph-based and AI-enhanced IoMT framework

Millions of people throughout the world struggle with mental health disorders, but the widespread stigma associated with these issues often prevents them from seeking treatment. We propose a novel strategy that integrates Internet of Medical Things (IoMT), DAG-based hedera technology, and Artificial Intelligence (AI) to overcome these challenges. We also consider the costs of chronic diseases such as Parkinson's and Alzheimer's, which often require 24-hour care. Using smart monitoring tools coupled with AI algorithms that can detect early indicators of deterioration, our system aims to provide low-cost, continuous support. Since IoMT data is large in volume, we need a blockchain network with high transaction throughput without compromising the privacy of patient data. To address this concern, we propose to use Hedera technology to ensure the privacy, and security of personal mental health information, scalability and a faster transaction confirmation rate. Overall, this research paper outlines a holistic approach to mental health monitoring that respects privacy, promotes accessibility, and harnesses the potential of emerging technologies. By combining IoMT, Hedera, and AI, we offer a solution that can help break down the barriers that prevent individuals from seeking the support they need for their mental well-being. Furthermore, comparative analysis shows that our best-performing ML models achieve an accuracy of around 98%, which is more than 30% better than traditional models such as logistic regression.

Enhancing machine learning thermobarometry for clinopyroxene-bearing magmas

In this study, we proposed a general workflow that aims to enhance the ML-based geothermobarometer modelling. Our workflow focuses on three key areas. Firstly, we developed a robust pre-processing pipeline that addresses data imbalance, feature engineering, and data augmentation. Secondly, we assessed modelling errors using a Monte Carlo approach to quantify the impact of analytical uncertainties on the final pressure and temperature estimates. Thirdly, we implemented a robust strategy to validate and test the ML models to avoid over- and under-fitting issues while correcting biases associated with the application of specific ML models (i.e., tree-based ensembles).To facilitate the use of our workflow, we have developed a web app (https://bit.ly/ml-pt-web) and a Python module (https://bit.ly/ml-pt-py). The robustness of this strategy has been tested on two calibrations: clinopyroxene (cpx) and clinopyroxene-liquid (cpx-liq). Our results show a significant reduction in errors compared to the baseline model, as well as good generalization ability on an independent external dataset. The Root Mean Squared Errors are 57 °C and 2.5 kbar for the cpx calibration, and 36 °C and 2.1 kbar for the cpx-liq calibration. Finally, our models show improved outcomes on the external dataset compared to existing ML and classical cpx and cpx-liq thermobarometers.

Machine learning approaches to predict the execution time of the meteorological simulation software COSMO

AbstractPredicting the execution time of weather forecast models is a complex task, since these models are usually performed on High Performance Computing systems that require large computing capabilities. Indeed, a reliable prediction can imply several benefits, by allowing for an improved planning of the model execution, a better allocation of available resources, and the identification of possible anomalies. However, to make such predictions is usually hard, since there is a scarcity of datasets that benchmark the existing meteorological simulation models. In this work, we focus on the runtime predictions of the execution of the COSMO (COnsortium for SMall-scale MOdeling) weather forecasting model used at the Hydro-Meteo-Climate Structure of the Regional Agency for the Environment and Energy Prevention Emilia-Romagna. We show how a plethora of Machine Learning approaches can obtain accurate runtime predictions of this complex model, by designing a new well-defined benchmark for this application task. Indeed, our contribution is twofold: 1) the creation of a large public dataset reporting the runtime of COSMO run under a variety of different configurations; 2) a comparative study of ML models, which greatly outperform the current state-of-practice used by the domain experts. This data collection represents an essential initial benchmark for this application field, and a useful resource for analyzing the model performance: better accuracy in runtime predictions could help facility owners to improve job scheduling and resource allocation of the entire system; while for a final user, a posteriori analysis could help to identify anomalous runs.

Model Complexity Reduction for ZKML Healthcare applications

Abstract Web 3.0 represents the next significant evolution of the internet that embodies the underlying decentralized network architectures, distributed ledgers, and advanced AI capabilities. Though the technologies are maturing rapidly, considerable barriers exist to high-scale adoption. The author discussed the barriers and the mitigations through specific technologies maturing to solve those issues. These include privacy-preserving technologies, off-chain and on-chain design optimizations, and the multi-dimensional approach needed in planning and adopting these technologies. As an extension, this paper discusses how one such enabler, ZKML, combines two streams of technology that are merging in unique ways to address problems in privacy and the cost of inference. The authors have conceptualized the technical and operational feasibility and implemented a reference healthcare implementation using the synthetic ICHOM in the evaluation phase in a global healthcare setting for high-volume data collection, including patient-reported outcomes. Model complexity reduction is researched and reported for the ICHOM diabetes dataset to advance the usage of ML models in global standards of healthcare data collection in network decentralized architectures for increased data protection and efficiencies.

Enhanced hydrological drought prediction in the Gediz Basin: integrating meteorological drought via hybrid wavelet-machine learning-random oversampling models using

ABSTRACT In study, meteorological drought ​​was used to estimate the potential hydrological drought that may occur in the Gediz Basin of Turkey. For this purpose, the most effective stream flow gauging station was determined by looking at the correlation values ​​between the meteorological data obtained from the Uşak meteorological station. SPEI values ​​for meteorological drought and SRI values ​​for hydrological drought are calculated for 3-, 6-, 9-, and 12-month periods. Correlation matrices were created between meteorological drought inputs from SPEI(t) to SPEI(t-12) and SRI(t) for use in hydrological drought models for 3-, 6-, 9- and 12-month periods. ML models were developed considering correlation matrices and it was seen that ML model results were not sufficient. For this reason, W-ML models were developed by applying DWT and Optuna hyperparameter analysis. It has been observed that the performance of W-ML models increases. Random oversampling (ROS), which has never been used in drought modeling, was then applied to W-ML models. W-ML-ROS model obtained an R2 value of 0.999 for testing set in 12-month period. Similarly, R2 values ​​for SRI3, SRI6 and SRI9 were obtained as 0.893, 0.851, and 0.940, respectively. Results showed that W-ML-ROS hybrid models can be used to predict hydrological drought from meteorological drought.

Development of a pig wean-quality score using machine-learning algorithms to characterize and classify groups with high mortality risk under field conditions

Mortality during the post-weaning phase is a critical indicator of swine production system performance, influenced by a complex interaction of multiple factors of the epidemiological triad. This study leveraged retrospective data from 1723 groups of pigs marketed within a US swine production system to develop a Wean-Quality Score (WQS) using machine learning techniques. The study evaluated three machine learning models, Random Forest, Support Vector Machine, and Gradient Boosting Machine, to classify groups having high or low 60-day mortality, where high mortality groups represented 25 % of the groups among the study population with the highest mortality values (n=431; 60-day mortality=9.98 %), and the remaining 75 % of the groups were of low mortality (n=1292; 60-day mortality=2.75 %). The best-performing model, Random Forest (RF), outperformed the other ML models in terms of accuracy (0.90), sensitivity (0.84), and specificity (0.92) metrics, and was then selected for further analysis, which consisted of creating the WQS and ranking the most important factors for classifying groups as high or low mortality. The most important factors ranked through the RF model to classify groups with high mortality were pre-weaning mortality, weaning age, average parity of litters in sow farms, and PRRS status. Additionally, stocking conditions such as stocking density and time to fill the barn were important predictors of high mortality. The WQS was developed and correlated (r = 0.74) with the actual 60-day mortality of the groups, offering a valuable tool for assessing post-weaning survivability in swine production systems before weaning. This study highlights the potential of machine learning and comprehensive data utilization to improve the assessment and management of weaned pig quality in commercial swine production, which producers can utilize to identify and intervene in groups, according to the WQS.

Continuous glucose monitoring using machine learning models and IoT device data: A meta-analysis.

Machine learning offers diverse options for effectively managing blood glucose levels in diabetes patients. Selecting the right ML algorithm is critical given the array of available choices. Integrating data from IoT devices presents promising opportunities to enhance real-time blood glucose management models. This meta-analysis aims to evaluate the effectiveness of machine learning models utilizing IoT device data for predicting blood glucose levels. We systematically searched electronic databases for studies published between 2019 and 2023. We excluded studies lacking ML model derivation or performance metrics. The Quality Assessment of Diagnostic Accuracy Studies tool assessed study quality. Our primary outcomes compared ML models for BG level prediction across different prediction horizons (PHs). We analyzed ten eligible studies across prediction horizons of 15, 30, 45, and 60 minutes. ML models exhibited mean absolute RMSE values of 15.02 (SD 1.45), 21.488 (SD 2.92), 30.094 (SD 3.245), and 35.89 (SD 6.4) mg/dL, respectively. Random Forest demonstrated superior performance across these PHs. We observed significant heterogeneity across all subgroups, indicating diverse sources of variability. As the PH lengthened, the RMSE for blood glucose prediction by the ML model increased, with Random Forest showing the highest relative performance among the ML models.

Predicting the efficiency of chidamide in patients with angioimmunoblastic T-cell lymphoma using machine learning algorithm.

Chidamide is subtype-selective histone deacetylase (HDAC) inhibitor that showed promising result in clinical trials to improve prognosis of angioimmunoblastic T-cell lymphoma (AITL) patients. However, in real world settings, contradictory reports existed as to whether chidamide improve overall survival (OS). Therefore, we aimed to develop an interpretable machine learning (Machine learning)-based model to predict the 2-year overall survival of AITL patients based on chidamide usage and baseline features. A total of 183 patients with AITL were randomly divided into training set and testing set. We used 5ML algorithms to build predictive models. Recursive feature elimination (RFE) method was used to filter for the most important features. The ML models were interpreted and the relevance of the selected features was determined using the Shapley additive explanations (SHAP) method and the local interpretable model-agnostic explanationalgorithm. A total of 183 patients with newly diagnosed AITL from 2012 to 2022 from 3 centers in China were enrolled in our study. Seventy-one patients were dead within 2years after diagnosis. Five ML algorithms were built based on chidamide usage and 16 baseline features to predict 2-year OS. Catboost model presented to be the best predictive model. After RFE screening, 12 variables demonstrated the best performance (AUC = 0.8651). Using chidamide ranked third among all the variables that correlated with 2-year OS. This study demonstrated that the Catboost model with 12 variables could effectively predict the 2-year OS of AITL patients. Combining chidamide in the treatment therapy was positively correlated with longer OS of AITL patients.

The hawk eye scan: Halyomorpha halys detection relying on aerial tele photos and neural networks

This paper faces the challenge of monitoring the Brown Marmorated Stink Bug (H.halys) (Halyomorpha halys) within orchards, utilizing drones, and computer vision. H.halys is an invasive species originating from East Asia, that is extremely polyphagous and poses a significant threat to various crops. Our first contribution is a drone navigation protocol, which ensures risk-free drone flights in cluttered orchard environments, preserving image quality and avoiding obstacles. We then create a pioneering H.halys dataset consisting of aerial telephotos captured in the field autonomously by the drone. The dataset allows the development and evaluation for the first time of multiple ML models for H.halys detection in the field. We trained YOLOv5, YOLOv8, RetinaNet, and Faster-RCNN models using different learning methodologies, exploiting different percentages of images without the bug, and using different slicing procedures for the images. The Medium YOLOv5 model trained with all images containing a bug detects the largest number of H.halys on the testing set and overall performs the best, while RetinaNet and Faster-RCNN provide the best trade-off between precision and recall. Models vary in their ability to handle occluded H.halys and bug-free images, which are common since the presence of the bug cannot be predicted before capturing a photo. These results show promising potential for automating H.halys monitoring, despite the image complexity and the early dataset stage. Our work marks a significant step towards enhancing smart agriculture practices due to the simplicity of the data acquisition process and the off-the-shelf hardware selection.

Developing effective optimized machine learning approaches for settlement prediction of shallow foundation

The precise assessment of shallow foundation settlement on cohesionless soils is a challenging geotechnical issue, primarily due to the significant uncertainties related to the factors influencing the settlement. This study aims to create an advanced hybrid machine learning methodology for accurately estimating shallow foundations' settlement (Sm). The initial contribution of the current research is developing and validating a robust hybrid optimization methodology based on an artificial electric field and single candidate optimizer (AEFSCO). This approach is thoroughly tested using various benchmark functions. AEFSCO will also be used to optimize three useful machine learning methods: long short-term memory (LSTM), support vector regression (SVR), and multilayer perceptron neural network (MLPNN) by adjusting their hyperparameters for predicting the settlement of shallow foundations. A database consisting of 189 individual case histories, conducted through various investigations, was used for training and testing the models. The database includes five input parameters and one output. These factors encompassed both the geometric characteristics of the foundation and the properties of the sandy soil. The results demonstrate that employing effective optimization strategies to adjust the ML models’ hyperparameters can significantly improve the accuracy of predicted results. The AEFSCO has increased the coefficient of determination (R2) value of the MLPNN model by 9.3 %, the SVR model by 8 %, and the LSTM model by 22 %. Also, the LSTM-AEFSCO model is more accurate than the SVR-AEFSCO and MLPNN-AEFSCO models. This is shown by the fact that R2 went from 0.9494 to 0.9290 to 0.9903, which is an increase of 4.5 % and 6 %.

Analyzing the heterogenous effects of factors on high-range speeding likelihood of taxi speeders: Does explainable deep learning provides more insights than random parameter approach?

The random parameters Generalized Linear Model (GLM) is frequently used to model speeding characteristics and capture the heterogenous effects of factors. However, this statistical approach is seldom employed for prediction and generalization due to the challenge of transferring its predefined errors. Recently, the emergence of explainable AI techniques has illuminated a new path for analyzing factors associated with risky driving behaviors. Despite this, there remains a gap that comparing results from machine and deep learning (ML/DL) approaches with those from random parameters GLM. This study aims to apply the random parameter GLM and explainable deep learning to evaluate the heterogenous effects of factors on the taxis’ high-range speeding likelihood. Initially, a Beta GLM with random parameters (BGLM-RP) is developed to model the high-range speeding likelihood among taxi drivers. Additionally, XGBoost, a simple convolutional neural network (Simple-CNN), a deeper CNN (DCNN), and a deeper CNN with self-attention (DCNN-SA) are developed. The quantified explanations and illustrations of the factors’ heterogenous effects from ML/DL models are derived from pseudo coefficients by decomposing factors’ SHapley Additive exPlanations (SHAP) values. All the developed statistical, ML, and DL models are compared in terms of mean absolute errors and mean square errors on testing and full data. Results show that DCNN-SA excels in prediction on testing data, indicating its superior generalization capabilities, while BGLM-RP outperforms other models on full data. The DCNN-SA can reveal the heterogenous effects of factors for both in-sample and out-of-sample data, which is not possible for the random parameter GLM. However, BGLM-RP can reveal larger magnitudes of the factors’ heterogenous effects for in-sample data. The signs and significances are identical between the varying coefficients from BGLM-RP and the pseudo coefficients from the ML/DL models, demonstrating the validity and rationale of using the proposed explanation framework to quantify the factors’ effects in ML/DL models. The study also discusses the contributions of various factors to the high-range speeding likelihood of taxi drivers.