Accuracy Of Machine Learning Models Research Articles

A comprehensive image dataset for the identification of lemon leaf diseases and computer vision applications.

A comprehensive dataset on lemon leaf disease can surely bring a lot of potentials into the development of agricultural research and the improvement of disease management strategies. This dataset was developed from 1354 raw images taken with professional agricultural specialist guidance from July to September 2024 in Charpolisha, Jamalpur, and further enhanced with augmented techniques, adding 9000 images. The augmentation process involves a set of techniques-flipping, rotation, zooming, shifting, adding noise, shearing, and brightening-to increase variety for different lemon leaf condition representations. Each of these images was standardized to 800 × 800 pixels resolution, so that consistency may be maintained among the dataset. All images were labelled in the nine prefixed categories: anthracnose, bacterial blight, citrus canker, curl virus, deficiency leaf, dry leaf, healthy leaf, sooty mould, and spider mites. In the present study, a DenseNet-121 architecture was used, where 20 % of the dataset was kept for validation and the remaining 80 % for training. A trained model with a batch size of 32 was trained for 30 epochs, achieving an accuracy of 98.56 % with augmentation, and 96.19 % without it. The dataset will not only act as a benchmark in developing accurate machine learning models for early disease detection, but it will also contribute to the cause of sustainable lemon cultivation practices by facilitating timely and effective disease management interventions.

Physics-Based Machine Learning to Predict Hydration Free Energies for Small Molecules with a Minimal Number of Descriptors: Interpretable and Accurate.

Hydration free energy (HFE) of molecules is a fundamental property having importance throughout chemistry and biology. Calculation of the HFE can be challenging and expensive with classical molecular dynamics simulation-based approaches. Machine learning (ML) models are increasingly being used to predict HFE. Although the accuracy of ML models for data sets for small molecules is impressive, these models suffer from lack of interpretability. In this work, we have developed a physics-based ML model with only six descriptors, which is both accurate and fully interpretable, and applied it to a database for small molecule HFE, FreeSolv. We evaluated the electrostatic energy by an approximate closed form of the Generalized Born (GB) model and polar surface area. In addition, we have log P and hydrogen bond acceptor and donors as descriptors along with the number of rotatable bonds. We have used different ML models, such as random forest and extreme gradient boosting. The best result from these models has a mean absolute error of only 0.74 kcal/mol. The main power of this model is that the descriptors have clear physical meaning, and it was found that the descriptor describing the electrostatics and the polar surface area, followed by the hydrogen bond donors and acceptors, are the most important factors for the calculation of hydration free energy.

Machine Learning for the Early Prediction of Delayed Cerebral Ischemia in Patients With Subarachnoid Hemorrhage: Systematic Review and Meta-Analysis.

Delayed cerebral ischemia (DCI) is a primary contributor to death after subarachnoid hemorrhage (SAH), with significant incidence. Therefore, early determination of the risk of DCI is an urgent need. Machine learning (ML) has received much attention in clinical practice. Recently, some studies have attempted to apply ML models for early noninvasive prediction of DCI. However, systematic evidence for its predictive accuracy is still lacking. The aim of this study was to synthesize the prediction accuracy of ML models for DCI to provide evidence for the development or updating of intelligent detection tools. PubMed, Cochrane, Embase, and Web of Science databases were systematically searched up to May 18, 2023. The risk of bias in the included studies was assessed using PROBAST (Prediction Model Risk of Bias Assessment Tool). During the analysis, we discussed the performance of different models in the training and validation sets. We finally included 48 studies containing 16,294 patients with SAH and 71 ML models with logistic regression as the main model type. In the training set, the pooled concordance index (C index), sensitivity, and specificity of all the models were 0.786 (95% CI 0.737-0.835), 0.77 (95% CI 0.69-0.84), and 0.83 (95% CI 0.75-0.89), respectively, while those of the logistic regression models were 0.770 (95% CI 0.724-0.817), 0.75 (95% CI 0.67-0.82), and 0.71 (95% CI 0.63-0.78), respectively. In the validation set, the pooled C index, sensitivity, and specificity of all the models were 0.767 (95% CI 0.741-0.793), 0.66 (95% CI 0.53-0.77), and 0.78 (95% CI 0.71-0.84), respectively, while those of the logistic regression models were 0.757 (95% CI 0.715-0.800), 0.59 (95% CI 0.57-0.80), and 0.80 (95% CI 0.71-0.87), respectively. ML models appear to have relatively desirable power for early noninvasive prediction of DCI after SAH. However, enhancing the prediction sensitivity of these models is challenging. Therefore, efficient, noninvasive, or minimally invasive low-cost predictors should be further explored in future studies to improve the prediction accuracy of ML models. PROSPERO (CRD42023438399); https://tinyurl.com/yfuuudde.

Discovery of robust and highly specific microbiome signatures of non-alcoholic fatty liver disease

BackgroundThe pathogenesis of non-alcoholic fatty liver disease (NAFLD) with a global prevalence of 30% is multifactorial and the involvement of gut bacteria has been recently proposed. However, finding robust bacterial signatures of NAFLD has been a great challenge, mainly due to its co-occurrence with other metabolic diseases.ResultsHere, we collected public metagenomic data and integrated the taxonomy profiles with in silico generated community metabolic outputs, and detailed clinical data, of 1206 Chinese subjects w/wo metabolic diseases, including NAFLD (obese and lean), obesity, T2D, hypertension, and atherosclerosis. We identified highly specific microbiome signatures through building accurate machine learning models (accuracy = 0.845–0.917) for NAFLD with high portability (generalizable) and low prediction rate (specific) when applied to other metabolic diseases, as well as through a community approach involving differential co-abundance ecological networks. Moreover, using these signatures coupled with further mediation analysis and metabolic dependency modeling, we propose synergistic defined microbial consortia associated with NAFLD phenotype in overweight and lean individuals, respectively.ConclusionOur study reveals robust and highly specific NAFLD signatures and offers a more realistic microbiome-therapeutics approach over individual species for this complex disease.E4koFqVmYng21Zwjp47efAVideo

Machine Learning for Predicting Primary Graft Dysfunction After Lung Transplantation: An Interpretable Model Study.

Primary graft dysfunction (PGD) develops within 72 h after lung transplantation (Lung Tx) and greatly influences patients' prognosis. This study aimed to establish an accurate machine learning (ML) model for predicting grade 3 PGD (PGD3) after Lung Tx. This retrospective study incorporated 802 patients receiving Lung Tx between July 2018 and October 2023 (640 in the derivation cohort and 162 in the external validation cohort), and 640 patients were randomly assigned to training and internal validation cohorts in a 7:3 ratio. Independent risk factors for PGD3 were determined by integrating the univariate logistic regression and least absolute shrinkage and selection operator regression analyses. Subsequently, 9 ML models were used to construct prediction models for PGD3 based on selected variables. Their prediction performances were further evaluated. Besides, model stratification performance was assessed with 3 posttransplant metrics. Finally, the SHapley Additive exPlanations algorithm was used to understand the predictive importance of selected variables. We identified 9 independent clinical risk factors as selected variables. Among 9 ML models, the random forest (RF) model displayed optimal performance (area under the curve [AUC] = 0.9415, sensitivity [Se] = 0.8972, specificity [Sp] = 0.8795 in the training cohort; AUC = 0.7975, Se = 0.7520, Sp = 0.7313 in the internal validation cohort; and AUC = 0.8214, Se = 0.8235, Sp = 0.6667 in the external validation cohort). Further assessments on calibration and clinical usefulness indicated the promising applicability of the RF model in PGD3 prediction. Meanwhile, the RF model also performed best in terms of risk stratification for postoperative support (extracorporeal membrane oxygenation time: P < 0.001, mechanical ventilation time: P = 0.006, intensive care unit time: P < 0.001). The RF model had the optimal performance in PGD3 prediction and postoperative risk stratification for patients after Lung Tx.

Machine learning algorithms for predictive modeling of dyslipidemia-associated cardiovascular disease risk in pregnancy: a comparison of boosting, random forest, and decision tree regression

BackgroundCardiovascular diseases (CVD) are major contributors to maternal mortality and morbidity during pregnancy and increased atherogenic index of plasma levels is associated with a higher risk of CVD and obesity.MethodsIn this study, we utilized three different machine learning algorithms (boosting, random forest, and decision tree regression) to predict dyslipidemia-associated cardiovascular disease using atherogenic index and lipid profile parameters based on a cross-sectional study datasets of 112 pregnant women aged between 15 and 49 conducted at Aminu Kano Teaching Hospital.ResultsThe results showed that random forest regression outperformed both boosting and decision tree regression, recording the lowest error criteria (MSE = 0.071 and RMSE = 0.266) for evaluating the model. These findings indicated that all the three algorithms have the potential to effectively model the data from atherogenic indices and lipid profile parameters but random forest and boosting were found to outperform decision tree models with respective R2 values of 0.95 and 0.92.ConclusionsOverall, the study highlights the accuracy of machine learning models (random forest, boosting, and decision trees) in predicting dyslipidemia-associated cardiovascular diseases and the findings could contribute to the development of effective strategies for the prevention and treatment of dyslipidemia-associated cardiovascular diseases.

Robust kernel extreme learning machines for postgraduate learning performance prediction.

In the context of graduate learning in China, mentors are the teachers with the highest frequency of contact and the closest relationships with postgraduate students. Nevertheless, a number of issues pertaining to the relationship between mentors and postgraduate students have emerged with increasing frequency in recent years, resulting in a notable decline in the quality of graduate education. In this paper, we investigate the influence of the relationship between mentors and postgraduate students on the postgraduate learning performance, with postgraduate students' admission motivation and learning pressure acting as moderating variables. In practice, outliers often appear during the data collection stage, and they have a significant impact on the convergence speed and prediction accuracy of machine learning models. In order to mitigate the impact of outliers, we propose a novel kernel extreme learning machine model that is robust to outliers and name it a robust kernel extreme learning machine (RK-ELM). The RK-ELM model can automatically detect any data that may be corrupted by uncertain disturbances, thereby enhancing the robustness and generalization ability of the model. We take 873 full-time postgraduate students from universities in Zhejiang Province, China as the research object, and then form a dataset based on the postgraduates' questionnaire results and their grade point averages in the current academic year. Experimental results show that: 1) RK-ELM is an effective model for predicting postgraduate learning performance; 2) The relationship between mentors and postgraduates has a significant impact on learning performance, but it cannot directly predict learning performance; 3) The combination of the relationship between mentors and postgraduates and enrollment motivation can be used to predict learning performance, where the former can predict learning performance by influencing learning pressure.

Predicting macular hole surgery outcomes: Integrating preoperative OCT features with supervised machine learning statistical models.

To evaluate various supervised machine learning (ML) statistical models to predict anatomical outcomes after macular hole (MH) surgery using preoperative optical coherence tomography (OCT) features. This retrospective study analyzed OCT data from idiopathic MH eyes at baseline and at 1-month post-surgery. The dataset was split 80:20 between training and testing. XLSTAT® statistical software (Lumivero, USA) was used to train different ML models on 10°CT parameters: prefoveal posterior cortical vitreous status, epiretinal membrane, intraretinal cysts, foveal retinal pigment epithelium hyperreflectivity, MH basal diameter, MH area (MHA), hole-forming factor, MH index, tractional hole index, and diameter hole index. The most effective statistical model was identified and was further assessed for accuracy, sensitivity, and specificity on a separate testing dataset. Six ML statistical models were trained on 33,260°CT data points from 3326°CT images of 308 operated MH (300 patients) eyes. Following training and internal validation, the random forest (RF) model achieved the highest accuracy (0.92), precision (0.94), recall (0.97), and F-score (0.96), and lowest misclassification rate. RF model identified the MHA index as the best predictor of post-surgical anatomical success. Following external testing, the RF model confirmed the highest accuracy and lowest misclassification rate (8.8%). ML-based statistical models can be used to predict MH status after surgery. The RF model was the most accurate ML model, and the MHA index was the best predictor of postoperative hole closure after surgery based on preoperative OCT parameters. These predictions may help with future surgical planning for MH patients.

Machine Learning Based Insights for Metal-Organic Frameworks Synthesis: A Comparative and Explainable Analysis of ZIF-8 Morphology

Metal-organic frameworks (MOFs) such as zeolitic imidazolate framework-8 (ZIF-8) are promising nanomaterials for various applications like drug delivery and energy storage. The efficacy of ZIF-8 in these applications highly depends on its morphology, including size and shape. However, understanding and controlling morphology during synthesis is challenging due to the complex interactions among synthesis conditions such as precursor concentration and reaction temperature. Traditional trial-and-error methods for morphology optimization are inefficient and cannot effectively account for the combined effects of conditions. Machine learning (ML) offers a powerful alternative for morphology prediction, which can accelerate the reverse engineering process to better understand how synthesis conditions affect morphology. Despite recent advances, developing accurate ML models and selecting the appropriate ones for specific applications remain a challenge. This study addresses these issues by experimentally investigating how variations in synthesis conditions, such as precursor concentrations, solvent properties, and temperature, affect ZIF-8 morphology. Using experimental data, this work further built and compared three ML models: Random Forest (RF), Support Vector Regressor (SVR), and Neural Network (NN). Among these, the NN model has the best performance in terms of R-squared and mean squared errors. These ML models provide insights into how synthesis conditions affect ZIF-8, thus setting the basis for future studies aimed at optimizing conditions and guiding more efficient manufacturing strategy to expand the applications of this versatile nanomaterial.

Artificial intelligence guided search for van der Waals materials with high optical anisotropy.

The exploration of van der Waals (vdW) materials, renowned for their unique optical properties, is pivotal for advanced photonics. These materials exhibit exceptional optical anisotropy, both in-plane and out-of-plane, making them an ideal platform for novel photonic applications. However, the manual search for vdW materials with giant optical anisotropy is a labor-intensive process unsuitable for the fast screening of materials with unique properties. Here, we leverage geometrical and machine learning (ML) approaches to streamline this search, employing deep learning architectures, including the recently developed Atomistic Line Graph Neural Network. Within the geometrical approach, we clustered vdW materials based on in-plane and out-of-plane birefringence values and correlated optical anisotropy with crystallographic parameters. The more accurate ML model demonstrates high predictive capability, validated through density functional theory and ellipsometry measurements. Experimental verification with 2H-MoTe2 and CdPS3 confirms the theoretical predictions, underscoring the potential of ML in discovering and optimizing vdW materials with unprecedented optical performance.

SmartSeg: machine learning infrastructure software for accelerating medical image segmentation in patient-specific applications

ABSTRACT Patient-specific applications in biomechanics and orthopaedics call for segmentation of volumetric medical images. This task has historically been performed manually or semi-automatically, which entails significant effort by trained experts. Computer vision algorithms based on machine learning have emerged as an effective tool for accelerating medical image processing tasks through automation. The current work describes fully no-code software which supports development of custom machine learning models for automatic segmentation of volumetric medical images. This cloud-integrated desktop application allows researchers and biomedical engineers to define segmentation algorithms, configure annotated datasets, and train personalised machine learning algorithms with cloud compute in just a few clicks. A presented case study demonstrates an ideal workflow in the software using femoral cartilage segmentation from magnetic resonance imaging as a representative use case. Models developed in the software demonstrated mean dice similarity coefficients of up to 0.886 on a test cohort, which is competitive with previously reported methods developed with significantly larger data sets. The mean surface difference between digitised models reconstructed from ground truth and predicted segmentation was 0.33 mm. Results suggest the described software enables creation of accurate machine learning models with limited engineering effort.

Development of an Optimized Neural Network Model using Hyperparameters Optimization for Electrical Load Prediction

Electrical load prediction has become essential to the efficient operation, control and management of modern electric power systems. Various machine learning prediction models have been developed for electrical load prediction, this includes Support Vector Regression, Fuzzy Logic and Neural Network (NN) modelling approaches. However, incorrect selection of model hyperparameters, which are parameters that affect the output of the prediction models, could result in low prediction accuracy of machine learning models. Hence, the development of an optimized NN model for electrical load prediction was presented in this study. Historical daily data of temperature, rainfall, relative humidity and windspeed for Osogbo, Nigeria was obtained from the National Aeronautics and Space Administration website; while electrical load data for the same location was collected from the Transmission Company of Nigeria. The data captured a period of five years (2017 to 2021). The NN models were developed with MATLAB R2022a software, and two hyperparameters, hidden layers and neuron counts, were optimized using the Bayesian optimization technique to enhance the quality of the models. The models were evaluated using mean absolute error (MAE), and root mean square error (RMSE). The MAE and RMSE for the non-optimized NN model were 6.5247 and 8.2725 respectively. Meanwhile, for the optimized NN model, the MAE and RMSE were 5.6571 and 7.4289 respectively. The obtained results show that the optimized NN performed better than the non-optimized NN models. Therefore, for more accurate load prediction, the method developed of this research is suggested for use by utility providers.

Accuracy of machine learning models for pre-diagnosis and diagnosis of pancreatic ductal adenocarcinoma in contrast-CT images: a systematic review and meta-analysis

Accuracy of machine learning models for pre-diagnosis and diagnosis of pancreatic ductal adenocarcinoma in contrast-CT images: a systematic review and meta-analysis

Machine Learning Prediction of the Corrosion Rate of Zinc-Based Alloys Containing Copper, Lithium, Magnesium, and Silver.

Implementation of machine learning (ML) techniques in materials science often requires large data sets. However, a proper choice of features and regression methods allows the construction of accurate ML models able to work with a relatively small data set. In this work, an extensive, although still limited, experimental data set of corrosion-related properties of Zn-based alloys used in biomedicine was created. On the basis of this data set, a robust and accurate model was built to predict the corrosion behavior of Zn-based alloys. This work highlights the effectiveness of ML methods for assessing the corrosion behavior of Zn-based alloys, which can facilitate their application in bioimplants.

Classifying continuous GNSS stations using integrated machine learning

The development of Global Navigation Satellite Systems (GNSS) results in large spatial geodetic networks with a distinct range of accuracy. Thus, classification of the GNSS stations is needed to determine which stations are appropriate for geodetic applications. Additionally, advanced Machine Learning (ML) techniques have been proposed. However, ML algorithms may sometimes be less sensitive due to a lack of samples or anomalies in input data. Therefore, this study introduces an approach in which human-based supervision is integrated into ML processes to improve the ML model’s performance in classifying the continuous GNSS stations. The human factor influences the ML processes through two sampling strategies: “suggest-decide” and “correct-retrain”, where the accuracy of ML models will be improved via human-based corrections. The idea is that the unsupervised ML-based clustering techniques are driven by human-based supervision to create samples for training the supervised ML-based classification models. In this study, we develop a MATLAB app to automate the clustering and labeling processes. Our finding demonstrates that applying these sampling strategies can enhance the accuracy of the ML-based classification models from under 50 % up to ∼99 % after re-training. Also, this study categorizes almost 9000 continuous monitoring stations in the Nevada database, of which 1900 stations in Europe serve as samples for training the ML-based classification models. Furthermore, the methodologies developed in this study can be applied to warning systems, which utilize internal and external human resources to correct errors, address unusual situations, and provide timely feedback for better performance of ML-based forecasts.

What can we learn from machine learning studies on flow diverter aneurysm embolization? A systematic review

BackgroundAs the use of flow diverters has expanded in recent years, predicting successful outcomes has become more challenging for certain aneurysms.ObjectiveTo provide neurointerventionalists with an understanding of the available machine...

Enhancing Power Conversion Efficiency of Perovskite Solar Cells Through Machine Learning Guided Experimental Strategies

AbstractPredicting the power conversion efficiency (PCE) using machine learning (ML) can effectively accelerate the experimental process of perovskite solar cells (PSCs). In this study, a high‐quality dataset containing 2079 experimental PSCs is established to predict PCE values using an accurate ML model, achieving an impressive coefficient of determination (R2) value of 0.76. In the 12 validation experiments with PSCs, the average absolute error between the observed and predicted PCE values is only 1.6%. Leveraging the recommended improvement solutions from the ML model, the device's PCE to 25.01% in experimental PSCs is successfully enhanced, thus truly realizing the objective of machine learning‐guided experiments. In addition, by improving the PCE of specific devices, the predicted value can reach 28.19%. The ML model has provided feasible strategies for experimentally improving the PCE of PSCs, which play a crucial role in achieving PCE breakthroughs.

Comparison of different machine learning methods in river streamflow estimation using isovel contours and hydraulic variables

ABSTRACT This study evaluates several machine learning (ML) models for estimating streamflow in the Osage River, Missouri, USA, and the Severn River, UK, using hydraulic input variables. These input variables are the flow section area (A), wetted perimeter (Pw ), water surface width (W), and velocity parameter (U) derived from isovel contours. Before using these models, the influential variables are selected using the sequential feature selection (SFS) method to reduce model complexity while maintaining performance. The two most important input variables identified were A and U, reducing the number of inputs from four to two. The results showed that the river’s flow conditions affect the accuracy of ML models used for estimating streamflow. Linear models such as MLR and ANFIS perform better in steady flow conditions (Osage River). In contrast, decision tree-based and non-linear models such as SVR with a radial basis kernel function (SVR_RBF) are better for unsteady flow conditions (Severn River). The finding suggests simpler models outperform complex deep-learning approaches (such as LSTM) for estimating streamflow by hydraulic variables. By selecting appropriate and efficient ML models, hydrometer stations can significantly improve the accuracy of streamflow estimation, leading to more informed decision-making in water management.

Using Artificial Intelligence to Detect Risk of Family Violence: Protocol for a Systematic Review and Meta-Analysis.

Despite the implementation of prevention strategies, family violence continues to be a prevalent issue worldwide. Current strategies to reduce family violence have demonstrated mixed success and innovative approaches are needed urgently to prevent the occurrence of family violence. Incorporating artificial intelligence (AI) into prevention strategies is gaining research attention, particularly the use of textual or voice signal data to detect individuals at risk of perpetrating family violence. However, no review to date has collated extant research regarding how accurate AI is at identifying individuals who are at risk of perpetrating family violence. The primary aim of this systematic review and meta-analysis is to assess the accuracy of AI models in differentiating between individuals at risk of engaging in family violence versus those who are not using textual or voice signal data. The following databases will be searched from conception to the search date: IEEE Xplore, PubMed, PsycINFO, EBSCOhost (Psychology and Behavioral Sciences collection), and Computers and Applied Sciences Complete. ProQuest Dissertations and Theses A&I will also be used to search the grey literature. Studies will be included if they report on human adults and use machine learning to differentiate between low and high risk of family violence perpetration. Studies may use voice signal data or linguistic (textual) data and must report levels of accuracy in determining risk. In the data screening and full-text review and quality analysis phases, 2 researchers will review the search results and discrepancies and decisions will be resolved through masked review of a third researcher. Results will be reported in a narrative synthesis. In addition, a random effects meta-analysis will be conducted using the area under the receiver operating curve reported in the included studies, assuming sufficient eligible studies are identified. Methodological quality of included studies will be assessed using the risk of bias tool in nonrandomized studies of interventions. As of October 2024, the search has not commenced. The review will document the state of the research concerning the accuracy of AI models in detecting the risk of family violence perpetration using textual or voice signal data. Results will be presented in the form of a narrative synthesis. Results of the meta-analysis will be summarized in tabular form and using a forest plot. The findings from this study will clarify the state of the literature on the accuracy of machine learning models to identify individuals who are at high risk of perpetuating family violence. Findings may be used to inform the development of AI and machine learning models that can be used to support possible prevention strategies. PROSPERO CRD42023481174; https://www.crd.york.ac.uk/prospero/display_record.php?RecordID=481174. PRR1-10.2196/54966.

Identify characteristics of Vietnamese oral squamous cell carcinoma patients by machine learning on transcriptome and clinical-histopathological analysis

Background/purposeOral squamous cell carcinoma (OSCC) is notorious for its low survival rates, due to the advanced stage at which it is commonly diagnosed. To enhance early detection and improve prognostic assessments, our study harnesses the power of machine learning (ML) to dissect and interpret complex patterns within mRNA-sequencing (RNA-seq) data and clinical-histopathological features. Materials and methods206 retrospective Vietnamese OSCC formalin-fixed paraffin-embedded (FFPE) tumor samples, of which 101 were subjected to RNA-seq for classification based on gene expression. Then, learning models were built based on clinical-histopathological data to predict OSCC subtypes and propose potential biomarkers for the remaining 105 samples. Results2 distinct groups of OSCC with different clinical-histopathological characteristics and gene expression. Subgroup 1 was characterized by severe histopathologic features with immune response and apoptosis signatures while subgroup 2 was denoted by more clinical/pathological features, cell division and malignant signatures. XGBoost and SVM (Support Vector Machine) models showed the best performance in predicting subtype OSCC. The study also proposed 12 candidate genes as potential biomarkers for OSCC subtypes (6/group). ConclusionThe study identified characteristics of Vietnamese OSCC patients through a combination of mRNA sequencing and clinical-histopathological analysis. It contributes to the insight into the tumor microenvironment of OSCC and provides accurate ML models for biomarker prediction using clinical-histopathological features.