Machine Learning Methods Research Articles

Innovative Machine Learning and Stakeholder Method to Assess Bycatch of Tursiops truncatus in the Southern Gulf of Mexico

The assessment and mitigation of bycatch, currently identified as the most significant threat to marine mammals, represents a substantial challenge for society. This issue is particularly acute in developing countries, where data on small-scale fisheries are scarce, and knowledge gaps exist regarding the distribution and abundance of various marine mammal species. Artisanal fisheries, particularly in developing countries, have been linked to significant mortality levels of marine organisms due to bycatch. The magnitude of this phenomenon reveals alarming figures. Notably, there is a high incidence of interactions between the bottlenose dolphin (Tursiops truncatus) and nearshore gillnets, where the overlap in their coastal distribution creates high-risk zones. The imperative to assess bycatch is driven not only by conservation principles but is also essential for sustainability in developing countries due to U.S. government regulations on imports of fishery products aimed at reducing bycatches worldwide. This study proposes an innovative methodology to investigate marine mammal bycatch in the southern Gulf of Mexico. This methodology is based on the development of artificial intelligence models, the integration of stakeholder input, and the use of habitat suitability models. This approach utilizes 11 years of sighting records and 1,654 spatial-temporal fishing effort data points collected through interviews with fishers. Additionally, the study develops artificial intelligence models, specifically Random Forest algorithms in Python, to enhance the analysis and prediction of bycatch risk. This research identified monthly variations in high-risk zones for marine mammal bycatch in the southern Gulf of Mexico, highlighting regions with a higher likelihood of interaction with gillnets. This pioneering work of applying artificial intelligence to marine mammal bycatch provides a complementary analysis for areas with limited economic and data resources.

TIDGN: A Transfer Learning Framework for Predicting Interactions of Intrinsically Disordered Proteins with High Conformational Dynamics.

Interactions between intrinsically disordered proteins (IDPs) are crucial for biological processes, such as intracellular liquid-liquid phase separation (LLPS). Experiments (e.g., NMR) and simulations used to study IDP interactions encounter a variety of difficulties, highlighting the necessity to develop relevant machine learning methods. However, reliable machine learning methods face the challenge resulting from the scarcity of available training data. In this work, we propose a transfer learning-based invariant geometric dynamic graph model, named TIDGN, for predicting IDP interactions. The model consists of a pretraining task module and a downstream task module. The pretraining task module learns the dynamic structural encoding of IDP monomers, which is then used by the downstream task module for interaction site prediction. The IDP monomer structure data set and the IDP interaction event data set are constructed using all-atom molecular dynamics (MD) simulations. The transfer learning strategy effectively enhances the model's performance. Both homotypic interactions and heterotypic interactions between two IDPs are considered in this work. Interestingly, TIDGN performs well for the heterotypic interaction prediction. Additionally, the feature ablation analysis emphasizes the importance of invariant geometric graph features. Taken together, our work demonstrates that the integration of transfer learning and the invariant geometric graph network offers a promising approach for addressing data scarcity challenges of IDP interaction prediction.

Predicting and Understanding Work Functions of Double Transition Metal MXenes via Interpretable Machine Learning Methods.

In this study, we employed interpretable machine learning models to predict and understand the work functions of double transition metal MXenes with the formula (M1)2(M2)X2O2. We created a comprehensive data set of 242 structures covering 11 transition metals and two X elements (C, N) using first-principles calculations. Various machine learning methods, including linear regression, support vector regression, random forest regression, and artificial neural networks, were thoroughly evaluated. The random forest model achieved the best performance on the test set, with a mean absolute error of 0.17 ± 0.02 eV, root-mean-square error of 0.24 ± 0.03 eV, and R2 of 0.86 ± 0.03. The feature importance analysis revealed a hierarchical influence mechanism: the properties of the outer transition metal (M1) dominate the work function, followed by the X element, while the inner transition metal (M2) has minimal impact. Furthermore, we employed the Sure Independence Screening and Sparsifying Operator method to derive analytical expressions that relate element features to work functions, achieving comparable accuracy (R2 = 0.82 ± 0.04) while providing physical insights. Our findings not only enable rapid prediction of MXene work functions but also provide valuable guidance for the rational design of MXene-based materials.

Generalizable, fast, and accurate DeepQSPR with fastprop

Quantitative Structure–Property Relationship studies (QSPR), often referred to interchangeably as QSAR, seek to establish a mapping between molecular structure and an arbitrary target property. Historically this was done on a target-by-target basis with new descriptors being devised to specifically map to a given target. Today software packages exist that calculate thousands of these descriptors, enabling general modeling typically with classical and machine learning methods. Also present today are learned representation methods in which deep learning models generate a target-specific representation during training. The former requires less training data and offers improved speed and interpretability while the latter offers excellent generality, while the intersection of the two remains under-explored. This paper introduces fastprop, a software package and general Deep-QSPR framework that combines a cogent set of molecular descriptors with deep learning to achieve state-of-the-art performance on datasets ranging from tens to tens of thousands of molecules. fastprop provides both a user-friendly Command Line Interface and highly interoperable set of Python modules for the training and deployment of feedforward neural networks for property prediction. This approach yields improvements in speed and interpretability over existing methods while statistically equaling or exceeding their performance across most of the tested benchmarks. fastprop is designed with Research Software Engineering best practices and is free and open source, hosted at github.com/jacksonburns/fastprop.

Revolutionizing sleep disorder diagnosis: A Multi-Task learning approach optimized with genetic and Q-Learning techniques

Adequate sleep is crucial for maintaining a healthy lifestyle, and its deficiency can lead to various sleep-related disorders. Identifying these disorders early is essential for effective treatment, which traditionally relies on polysomnogram (PSG) tests. However, diagnosing sleep disorders with high accuracy based solely on electroencephalogram (EEG) signals, rather than using various signals in a complex PSG, can reduce the time and cost required, and the need for specialized signal devices, as well as increase accessibility and usability. Previous studies have focused on traditional machine learning (ML) methods such as K-Nearest Neighbors (KNNs), Support Vector Machines (SVMs), and ensemble learning methods for sleep disorders analysis. However, these models require manual methods for feature extraction, and the prediction accuracy greatly depends on the type of feature extracted. Additionally, the EEG signal datasets are small and heterogeneous, challenging traditional machine learning and deep learning models. The study proposes an innovative multi-task learning convolutional neural network with a partially shared structure that uses frequency-time images generated from EEG signals to address these limitations. The proposed technique makes two predictions using non-shared features from time-frequency images created through Short Time Fourier Transform (STFT) and Continuous Wavelet Transform (CWT), one prediction from shared features, and the final prediction is a combination of these three predictions. The weights for this combination were optimized using the genetic algorithm and the Q-learning algorithm, aiming to minimize loss and maximize accuracy. The study utilizes a dataset involving 26 participants to examine the impact of Partial Sleep Deprivation (PSD) on EEG recordings. The outcomes demonstrated that the multi-task learning model using these two optimization methods, attained 98% accuracy on the test data for predicting partial sleep deprivation. This automated diagnostic model is an efficient supporting tool for rapidly and effectively diagnosing sleep disorders. It swiftly and precisely evaluates sleep data, minimizing the time and effort required by the patient and the physician.

Development of macrophage M2 relate signature for predicting prognosis and immunotherapy response in ovarian cancer

BackgroundOvarian cancer ranks as the fifth most common cause of cancer-related deaths in women worldwide. Macrophages M2 is believed to support tumor growth by suppressing immune responses and promoting angiogenesis.MethodsA macrophage M2-related signature (MRS) was developed by applying machine learning methods that included 10 algorithms and utilized data from the TCGA, GSE14764 and GSE140082 datasets. The predictive capacity of the MRS for immunotherapy response was evaluated through various methods, including immunophenoscore, TIDE score, TMB score, immune escape score, as well as two immunotherapy cohorts (IMvigor210 and GSE91061).ResultsThe optimal MRS, developed using the lasso algorithm, served as an independent prognostic factor and demonstrated stable performance in predicting overall survival rates in ovarian cancer. In the TCGA dataset, the AUC values for the 1-, 3-, and 5-year ROC curves were 0.874, 0.808, and 0.813, respectively. The C-index of our MRS was superior to that of clinical stage, tumor grade, and several other established prognostic signatures. Ovarian cancer patients with low risk score exhibited higher ESTIMATE score, increased levels of immune cells, elevated PDI&CTLA4 immunophenoscore, higher TMB score, reduced TIDE score, and lower immune escape score. Additionally, the survival prediction nomogram displayed significant potential for clinical application in estimating the 1-, 3-, and 5-year overall survival rates of ovarian cancer patients.ConclusionOur study developed a novel MRS for ovarian cancer, which could act as an indicator for predicting the prognosis and immunotherapy response in ovarian cancer.

Identifying Lactylation-related biomarkers and therapeutic drugs in ulcerative colitis: insights from machine learning and molecular docking

BackgroundUlcerative colitis (UC), a chronic relapsing-remitting inflammatory bowel disease. Recent studies have shown that lactylation modifications may be involved in metabolic-immune interactions in intestinal inflammation through epigenetic regulation, but their specific mechanisms in UC still require in-depth validation.MethodsWe conducted comparative analyses of transcriptomic profiles, immune landscapes, and functional pathways between UC and normal cohorts. Lactylation-related differentially expressed genes were subjected to enrichment analysis to delineate their mechanistic roles in UC. Through machine learning algorithms, the diagnostic model was established. Further elucidating the mechanisms and regulatory network of the model gene in UC were GSVA, immunological correlation analysis, transcription factor prediction, immunofluorescence, and single-cell analysis. Lastly, the CMap database and molecular docking technology were used to investigate possible treatment drugs for UC.ResultsTwenty-two lactylation-related differentially expressed genes were identified, predominantly enriched in actin cytoskeleton organization and JAK-STAT signaling. By utilizing machine learning methods, 3 model genes (S100A11, IFI16, and HSDL2) were identified. ROC curves from the train and test cohorts illustrate the superior diagnostic value of our model. Further comprehensive bioinformatics analyses revealed that these three core genes may be involved in the development of UC by regulating the metabolic and immune microenvironment. Finally, regorafenib and R-428 were considered as possible agents for the treatment of UC.ConclusionThis study offers a novel strategy to early UC diagnosis and treatment by thoroughly characterizing lactylation modifications in UC.

Evaluating the factors influencing accuracy, interpretability, and reproducibility in the use of machine learning classifiers in biology to enable standardization

The complexity and variability of biological data has promoted the increased use of machine learning methods to understand processes and predict outcomes. These same features complicate reliable, reproducible, interpretable, and responsible use of such methods, resulting in questionable relevance of the derived. outcomes. Here we systematically explore challenges associated with applying machine learning to predict and understand biological processes using a well- characterized in vitro experimental system. We evaluated factors that vary while applying machine learning classifers: (1) type of biochemical signature (transcripts vs. proteins), (2) data curation methods (pre- and post-processing), and (3) choice of machine learning classifier. Using accuracy, generalizability, interpretability, and reproducibility as metrics, we found that the above factors significantly mod- ulate outcomes even within a simple model system. Our results caution against the unregulated use of machine learning methods in the biological sciences, and strongly advocate the need for data standards and validation tool-kits for such studies.

Machine Learning Algorithms of Remote Sensing Data Processing for Mapping Changes in Land Cover Types over Central Apennines, Italy

This work presents the use of remote sensing data for land cover mapping with a case of Central Apennines, Italy. The data include 8 Landsat 8-9 Operational Land Imager/Thermal Infrared Sensor (OLI/TIRS) satellite images in six-year period (2018–2024). The operational workflow included satellite image processing which were classified into raster maps with automatically detected 10 classes of land cover types over the tested study. The approach was implemented by using a set of modules in Geographic Resources Analysis Support System (GRASS) Geographic Information System (GIS). To classify remote sensing (RS) data, two types of approaches were carried out. The first is unsupervised classification based on the MaxLike approach and clustering which extracted Digital Numbers (DN) of landscape feature based on the spectral reflectance of signals, and the second is supervised classification performed using several methods of Machine Learning (ML), technically realised in GRASS GIS scripting software. The latter included four ML algorithms embedded from the Python’s Scikit-Learn library. These classifiers have been implemented to detect subtle changes in land cover types as derived from the satellite images showing different vegetation conditions in spring and autumn periods in central Apennines, northern Italy.

Risk of overestimating odor control performance with the removal efficiency of a single odorant: Odorant interactions during the chemical absorption

ABSTRACT Odor is a serious issue for municipal solid waste treatment process, and chemical absorption is a common technology for odor control. Interactions among odorants will influence the performance of chemical absorption, and they are still not fully understood. This work took the common and important odorants, including hydrogen sulfide (H2S), methanethiol, propanethiol and acetaldehyde, as examples, to investigate the removal performance of mixed odorants by sodium hypochlorite (NaClO) solution at different concentrations, and interactions among the odorants. The absorption experiments were conducted in gas-washing bottles with single or mixture of H2S, methanethiol, propanethiol and acetaldehyde as the inlet gases, and NaClO solutions at different concentrations as the absorption solutions. The thermodynamic equilibrium was simulated. Acetaldehyde was eliminated mainly by physical absorption, and the removal efficiency was not affected by the other three odorants. The removal efficiencies of H2S, methanethiol, and propanethiol increased with the chlorine concentration ([Cl2]), and reached nearly 100% by the NaClO solution of pH = 12.19, [Cl2] = 158.00 mg/L. H2S, methanethiol, and propanethiol competed for reacting with NaClO. H2S was more effectively removed than methanethiol and propanethiol due to its lower pKa value. The removal efficiency of methanethiol decreased linearly with the increase in methanethiol and H2S concentrations mainly due to the consumption of NaClO. Propanethiol removal was decreased by both methanethiol and H2S, and methanethiol had more influence than H2S due to the higher consumption of NaClO. The odor control performance could be overestimated when there are several important odorants, and only the removal efficiency of a single odorant was considered. Correspondingly, suggestions for chemical scrubber operation were provided, including the consideration of odorant interactions, the selection of monitoring odorants, and the optimization of operating parameters (pH and [Cl2]) using machine learning methods. Implications Chemical absorption is widely applied for odor control, and the interaction between odorants is an important influencing factor of the performance. Hydrogen sulfide, methanethiol, propanethiol and acetaldehyde are common and important odorants emitted during municipal solid waste treatment. This work investigated the removal performance of chemical absorption for the single and mixture of these odorants, and revealed the interaction between them, as well as the risk of overestimating odor performance with the removal efficiency of a single odorant, which can provide insights into optimizing odor control technologies.

AI Based Acoustic Wave Monitoring of Rail Defects

ABSTRACT: Rail is one of the most energy efficient and economical modes of transportation. Regular rail- way track health inspection is an essential part of a robust and secure train operation. Delayed investigations and problem discoveries pose a serious risk to th e safe functioning of rail transportation. The traditional method of manually examining the rail track using a railway cart is both inefficient and susceptible to mistakes and biasness. It is imperative to automate inspection in order to avert catastrophes and save countless lives, particularly in zones where train accidents are numerous. This research develops an Internet of Things (IoT)-based autonomous railway track fault detection scheme to enhance the existing railway cart system to address the aforementioned issues. In addition to data collection on Pakistani railway lines, this work contributes significantly to railway track fault identification and classification based on acoustic analysis, as well as fault localization. Based on their frequency of occurrences, six types of track faults were first targeted: wheel burnt, loose nuts and bolts, crash sleeper, creep, low joint, and point and crossing. Support vector machines, logistic regression, random forest, extra tree classifier, decision tree classifier, multilayer perceptron and ensemble with hard and soft voting were among the machine learning methods used. The results indicate that acoustic data can successfully assist in discriminating track defects and localizing these defects in real time. The results show that MLP achieved the best results, with an accuracy of 98.4 percent.

Prediction of KRAS gene mutations in colorectal cancer using a CT-based radiomic model

BackgroundDetermining the KRAS gene mutation status in colorectal cancer (CRC) before surgery is highly important for an individualized clinical treatment. This study aimed to explore the clinical value of radiomics models based on CT images in predicting the KRAS mutation status in patients with CRC.MethodsA total of 201 CRC patients who underwent surgery and pathology examinations from March 2022 to January 2025 were included. They were randomly allocated to a training group (160 patients) or a testing group (41 patients) at a ratio of 8:2. All patients underwent plain CT and contrast-enhanced examinations before surgery. The 3D segmentation of the tumour was manually delineated by two radiologists who were unaware of the pathological results and KRAS gene detection outcomes. The PyRadiomics package in Python was used to extract 2,264 radiomic features from each ROI. After dimensionality reduction, machine learning methods such as extremely randomized trees (ERT), random forest (RF), XGBoost, Bagging, and CatBoost were used for model construction. The performance of the models was compared using the area under the receiver operating characteristic curve (AUC), accuracy, sensitivity, and specificity. The Delong test was employed to assess the differences between the various models.ResultsAfter feature selection, the top 8 features with the highest mutual information scores were extracted to construct a prediction model. The Delong test revealed that the XGBoost model, which is based on CT images from the vein phase, performed the best, with AUC values of 0.90 and 0.81 in the training and test sets, respectively. The calibration curve indicated a high consistency between the actual and predicted probabilities of the samples. The decision curve analysis results revealed that the XGBoost model exhibited the highest net clinical benefit among all the models.ConclusionIn this study, a highly accurate radiomics model was developed for KRAS gene mutation status prediction in patients with CRC before surgery. This technique avoids the potential risks of tumour rupture and dissemination during biopsy and can serve as a powerful tool to assist doctors in developing personalized and precise targeted treatments for colorectal cancer, which highly important in clinical work.

How AI policies influence urban innovation in China: analysis based on feature extraction and fsQCA

PurposeThis study aims to understand the patterns that characterize the impact of artificial intelligence (AI) policies on urban innovation performance, and reveal how these patterns vary across different regions, thereby helping AI policy-making and promoting the urban innovation.Design/methodology/approachThis research focuses on how AI policies influence innovation using the city as unit of analysis. AI policy and patent data were collected from 156 Chinese cities over a decade. Coding and machine learning methods were applied to extract policy features, including three types of policy instruments, policy continuity, policy intensity, and policy count. The fuzzy set Qualitative Comparative Analysis (fsQCA) method is used to identify patterns that explain how AI policies influence urban innovation performance and to further explore regional differences.FindingsComparing four models for extracting policy instruments, ERNIE 3.0 has been proven to be the most accurate and effective model. Three patterns are found using fsQCA: the environment-safeguard, demand-pull, and supply-environment-demand triple-drive patterns. Moreover, these patterns reflect the development distinction of the eastern, middle, and western cities, respectively. Hence, governments should focus on the intricate interplay and synergistic application of multiple policy levers, and enhance creativity in policy formulation based on their specific developmental characteristics.Originality/valueThis research analyzed the patterns that AI policies influence urban innovation from the national and regional perspective. Automated methods were introduced for policy feature extraction, particularly in identifying policy instruments, thereby significantly cutting down on labor and enhancing the efficiency of data analysis. Besides, this research highlights the interplay among various factors, utilizing fsQCA to reveal the collaborative dynamics at work, which compensates for the deficiency of independent assumptions in regression analysis, and analyze the synergistic effects of different factors from a systematic perspective.

The potential of radiomics features in the detection of hepatocellular carcinoma (HCC) in 2D liver MRI images by using machine learning methods

The potential of radiomics features in the detection of hepatocellular carcinoma (HCC) in 2D liver MRI images by using machine learning methods

RGBChem: Image-Like Representation of Chemical Compounds for Property Prediction.

In this work, we introduce RGBChem, a novel approach for converting chemical compounds into image representations, which are subsequently used to train a convolutional neural network (CNN) to predict the HOMO-LUMO gap for compounds from the QM9 database. By modifying the arbitrary order of atoms present in .xyz files used to generate these images, it has been demonstrated that expanding the initial training set size can be achieved by creating multiple unique images (data points) from a single molecule. This study shows that the presented approach leads to a statistically significant improvement in model accuracy, highlighting RGBChem as a powerful approach for leveraging machine learning (ML) in scenarios where the available data set is too small to apply ML methods effectively.

Twitter Sentiment Analysis Using NLP

Abstract—Social media platforms, particularly Twitter, have become pivotal sources of public opinion and sentiment expression. The analysis of these sentiments has significant applications across various domains, including marketing, politics, and public health. This paper presents a comprehensive review and implementation of Natural Language Processing (NLP) techniques for sentiment analysis of Twitter data. We explore various preprocessing methods, feature extraction techniques, and machine learning algorithms specifically optimized for the unique characteristics of Twitter content. Our implementation demonstrates a pipeline that handles the challenges of Twitter data, including abbreviated language, emoticons, hashtags, and context-specific jargon. Using a large-scale dataset of 1.8 million tweets from Kaggle, our hybrid approach combining traditional machine learning methods with deep learning techniques achieves superior performance with an accuracy of 87.6%, an F1-score of 0.862, and significantly improved handling of negation and sarcasm compared to baseline methods. The analysis further reveals important insights into the temporal and contextual nature of sentiment expression on Twitter and suggests promising directions for future research in this domain.) Key Words:. sentiment analysis, natural language processing, Twitter, social media analytics, machine learning, deep learning, text classification

Pan-cancer predictive survival model development and evaluation using electronic health record and genetic data across 10 cancer types

The growing burden of cancer and recent surge in healthcare data availability call for new ways of analysing this multifactorial disease and improving patient outcomes. The aim of this study is to develop and evaluate prognostic cancer survival models across ten common cancer types based on a large patient sample. We compare the performance of different machine learning algorithms and assess the added value of genetic information in cancer prognosis. We also provide ways to improve model explainabilty which is critical for model adoption in clinical practice. This study included data from 9977 patients with bladder, breast, colorectal, endometrial, glioma, leukaemia, lung, ovarian, prostate, and renal cancers. Genetic data collected through the 100,000 Genomes Project was linked with clinical and demographic data provided by the National Cancer Registration and Analysis Service, Hospital Episode Statistics and Office for National Statistics. More than 500 prognostic features were assessed and four machine learning algorithms including Elastic Net Cox proportional hazards regression, random survival forest, gradient boosting survival and DeepSurv neural network were developed in this study. Most models achieved good performance varying from 60% in bladder cancer to 80% in glioma with the average C-index of 72% across all cancer types. Different machine learning methods achieved similar performance with DeepSurv model slightly underperforming compared to other methods. Addition of genetic data improved performance in endometrial, glioma, ovarian and prostate cancers, showing its potential importance for cancer prognosis. Patient’s age, stage, grade, referral route, waiting times, pre-existing conditions, previous hospital utilisation, tumour mutational burden and mutations in gene TP53 were among the most important features in cancer survival modelling. By offering a comprehensive set of predictive models for cancer survival, this study fills a critical gap in our understanding of cancer prognosis and provides new tools for informing cancer treatment and consequently improving patient outcomes.

Identification of predictive subphenotypes for clinical outcomes using real world data and machine learning

Predicting treatment response is an important problem in real-world applications, where the heterogeneity of the treatment response remains a significant challenge in practice. Unsupervised machine learning methods have been proposed to address this challenge by clustering patients with similar electronic health record (EHR) data. However, they cannot guarantee coherent outcomes within the groups. Here, we propose Graph-Encoded Mixture Survival (GEMS) as a general machine learning framework to identify distinct predictive subphenotypes that guarantee coherent survival and baseline characteristics within each subphenotype. We apply our method to a real-world dataset of advanced non-small cell lung cancer (aNSCLC) patients receiving first-line immune checkpoint inhibitor (ICI) therapy to predict overall survival (OS). Our method outperforms baseline methods for predicting OS and identifies three reproducible subphenotypes associated with distinct baseline clinical characteristics and OS. Our results demonstrate that our method can provide insights in the heterogeneity of treatment response and potentially influence treatment selection.

Peer Review of “The Order in Speech Disorder: A Scoping Review of State of the Art Machine Learning Methods for Clinical Speech Classification (Preprint)”

Peer Review of “The Order in Speech Disorder: A Scoping Review of State of the Art Machine Learning Methods for Clinical Speech Classification (Preprint)”

“Dynamic Cutting Force Estimation via Fourier Neural Operator (FNO) with Inferred 1 Machine Tool Dynamics: A Proof of Concept”

Abstract Knowing machine tool dynamics and cutting force is critical to machining process optimization, tool life prediction, and in-process monitoring. Identifying system dynamics and estimating dynamic cutting forces often require dedicated procedures and extensive execution effort. This paper presents a Fourier Neural Operator (FNO)-inspired end-to-end architecture for rapid system inference and dynamic force estimations through interpretable operator learning in the frequency domain using a set of system excitation and response data. This machine learning method extracts the system frequency response function (FRF) and produces a function map from acceleration to dynamic force. For validation, both a numerical study with a theoretical two-degree-of-freedom model and a field experiment on an actual machine tool are conducted. Results demonstrate that this FNO-based method rapidly determines the FRF at the tooltip and predicts dynamic forces with over 90% accuracy in terms of R-squared value for both validation cases. Model training considerations, limitations, and practicality of this method are also discussed in this paper.