Domain Experts Research Articles

Impact of generative artificial intelligence models on the performance of citizen data scientists in retail firms

Generative Artificial Intelligence (AI) models serve as powerful tools for organizations aiming to integrate advanced data analysis and automation into their applications and services. Citizen data scientists—individuals without formal training but skilled in data analysis—combine domain expertise with analytical skills, making them invaluable assets in the retail sector. Generative AI models can further enhance their performance, offering a cost-effective alternative to hiring professional data scientists. However, it is unclear how AI models can effectively contribute to this development and what challenges may arise. This study explores the impact of generative AI models on citizen data scientists in retail firms. We investigate the strengths, weaknesses, opportunities, and threats of these models. Survey data from 268 retail companies is used to develop and validate a new model. Findings highlight that misinformation, lack of explainability, biased content generation, and data security and privacy concerns in generative AI models are major factors affecting citizen data scientists’ performance. Practical implications suggest that generative AI can empower retail firms by enabling advanced data science techniques and real-time decision-making. However, firms must address drawbacks and threats in generative AI models through robust policies and collaboration between domain experts and AI developers.

Evaluating outlier probabilities: assessing sharpness, refinement, and calibration using stratified and weighted measures

An outlier probability is the probability that an observation is an outlier. Typically, outlier detection algorithms calculate real-valued outlier scores to identify outliers. Converting outlier scores into outlier probabilities increases the interpretability of outlier scores for domain experts and makes outlier scores from different outlier detection algorithms comparable. Although several transformations to convert outlier scores to outlier probabilities have been proposed in the literature, there is no common understanding of good outlier probabilities and no standard approach to evaluate outlier probabilities. We require that good outlier probabilities be sharp, refined, and calibrated. To evaluate these properties, we adapt and propose novel measures that use ground-truth labels indicating which observation is an outlier or an inlier. The refinement and calibration measures partition the outlier probabilities into bins or use kernel smoothing. Compared to the evaluation of probability in supervised learning, several aspects are relevant when evaluating outlier probabilities, mainly due to the imbalanced and often unsupervised nature of outlier detection. First, stratified and weighted measures are necessary to evaluate the probabilities of outliers well. Second, the joint use of the sharpness, refinement, and calibration errors makes it possible to independently measure the corresponding characteristics of outlier probabilities. Third, equiareal bins, where the product of observations per bin times bin length is constant, balance the number of observations per bin and bin length, allowing accurate evaluation of different outlier probability ranges. Finally, we show that good outlier probabilities, according to the proposed measures, improve the performance of the follow-up task of converting outlier probabilities into labels for outliers and inliers.

A dynamic graph structural framework for implicit sentiment identification based on complementary semantic and structural information

Implicit sentiment identification has become the classic challenge in text mining due to its lack of sentiment words. Recently, graph neural network (GNN) has made great progress in natural language processing (NLP) because of its powerful feature capture ability, but there are still two problems with the current method. On the one hand, the graph structure constructed for implicit sentiment text is relatively single, without comprehensively considering the information of the text, and it is more difficult to understand the semantics. On the other hand, the constructed initial static graph structure is more dependent on human labor and domain expertise, and the introduced errors cannot be corrected. To solve these problems, we introduce a dynamic graph structure framework (SIF) based on the complementarity of semantic and structural information. Specifically, for the first problem, SIF integrates the semantic and structural information of the text, and constructs two graph structures, structural information graph and semantic information graph, respectively, based on specialized knowledge, which complements the information between the two graph structures, provides rich semantic features for the downstream identification task, and helps to understanding of the contextual information between implicit sentiment semantics. To deal with the second issue, SIF dynamically learns the initial static graph structure to eliminate the noise information in the graph structure, preventing noise accumulation that affects the performance of the downstream identification task. We compare SIF with mainstream natural language processing methods in three publicly available datasets, all of which outperform the benchmark model. The accuracy on the Puns of day dataset, SemEval-2021 task 7 dataset, and Reddit dataset reaches 95.73%, 85.37%, and 65.36%, respectively. The experimental results demonstrate a good application scenario for our proposed method on implicit sentiment identification tasks.

Enhancing transparency in public procurement: A data-driven analytics approach

Open data is a strategy used by governments to promote transparency and accountability in public procurement processes. To reap the benefits of open data, exploring and analyzing the data is necessary to gain meaningful insights into procurement practices. However, accessing, processing, and analyzing open data can be challenging for non-data-savvy users with domain expertise, creating a barrier to leveraging open procurement data. To address this issue, we present the design, development, and implementation of a visual analytics tool. This tool automates data extraction from multiple sources, performs data cleansing, standardization, and database processing, and generates meaningful visualizations to streamline public procurement analysis. In addition, the tool estimates and visualizes corruption risk indicators at different levels (e.g., regions or public entities), providing valuable insights into the integrity of the procurement process. Key contributions of this work include: (1) providing a comprehensive guide to the development of an open procurement data visualization tool; (2) proposing a data pipeline to support processing, corruption risk estimator and data visualization; (3) demonstrating through a case study how visual analytics can effectively use open data to generate insights that promote and enhance transparency.

Blockchain-enabled architecture for lead acid battery circularity

Widespread use of lead acid batteries (LABs) is resulting in the generation of million tons of battery waste, globally. LAB waste contains critical and hazardous materials, which have detrimental effects on the environment and human health. In recent times, recycling of the LABs has become efficient but the collection of batteries in developing countries is not efficient, which led to the non-professional treatment and recycling of these batteries in the informal sector. This paper proposes a blockchain-enabled architecture for LAB circularity, which ensures authentic, traceable and transparent system for collection and treatment of batteries. The stakeholders—battery manufacturers, distributors, retailers, users, and validators (governments, domain experts, third party experts, etc.)—are integrated in the circular loop through a blockchain network. A mobile application user interface is provided to all the stakeholders for the ease of adoption. The batteries manufactured and supplied in a geographical region as well as the recycled materials at the battery end-of-life are traced authentically. This architecture is expected to be useful for the battery manufacturers to improve their extended producer responsibility and support responsible consumption and production.

Neurofibroma detection using joint imaging genomics and ensemble learning

Accurate detection of neurofibromas is crucial for quantitative monitoring of tumor progression and surgical assessment. This paper proposes a method for neurofibroma detection in whole-body magnetic resonance imaging (WBMRI) using joint imaging genomics and ensemble learning. Firstly, we enhance texture features through a combination of image sharpening, filtering, brightness adjustment, and contrast enhancement. Then, we employ a weighted boxes fusion (WBF) technique based on test-time augmentation (TTA) under a single model and further integrate multiple models using the dual fusion approach of TTA and WBF. For segmentation, we utilize minimum bounding boxes based on segmentation masks for position calibration. Finally, false positive tumor regions are further eliminated through imaging genomics features. The experimental MRI data is obtained from collaboration between Harvard Medical School and domestic tertiary hospitals, comprising 158 cases with a total of 1380 tumors. Five-fold cross-validation is conducted with segmentation annotations completed by domain experts. Compared to the best results of single models, our proposed method achieves a 10.1% increase in average precision (AP), 7.8% increase in sensitivity, reduction of average false positives to 3.58, a decrease of 17.68, and an 8.5% improvement in competitive performance metric (CPM). This method effectively enhances the accuracy of neurofibroma detection and is applicable to detecting tumors and lesions in other medical imaging applications.

Rule-Based DSL for Continuous Features and ML Models Selection in Multiple Sclerosis Research

Machine learning (ML) has emerged as a powerful tool in multiple sclerosis (MS) research, enabling more accurate diagnosis, prognosis prediction, and treatment optimization. However, the complexity of developing and deploying ML models poses challenges for domain experts without extensive programming knowledge. We propose a novel domain-specific language (DSL) that simplifies the process of selecting features, choosing appropriate ML models, and defining training rules for MS research. The DSL offers three approaches: AutoML for automated model and feature selection, manual selection for expert-guided customization, and a customizable mode allowing for fine-grained control. The DSL was implemented and evaluated using real-world MS data. By establishing task-specific DSLs, we have successfully identified workflows that enhance the filtering of ML models and features. This method is crucial in determining the T2-related MRI features that accurately predict both process speed time and walk speed. We assess the effectiveness of using our DSL to enhance ML models and identify feature importance within our private data, aiming to reveal the relationships between features. The proposed DSL empowers domain experts to leverage ML in MS research without extensive programming knowledge. By integrating MLOps practices, it streamlines the ML lifecycle, promoting trustworthy AI through explainability, interpretability, and collaboration. This work demonstrates the potential of DSLs in democratizing ML in MS and paves the way for future research in adaptive and evolving DSL architectures.

Screen Use Time and Its Association With Mental Health Issues in Young Adults in India: Protocol for a Cross-Sectional Study.

Screen use time has increased in the past decade owing to the increased availability and accessibility of digital devices and the internet. Several studies have shown an association between increased screen use time and mental health issues such as anxiety and depression. However, studies in the young adult population-a demographic with high screen use-and in low- and middle-income country settings are limited. This protocol describes a study that aims to measure self-reported screen use times and patterns in young adults (18-24 y) in India and assess if increased screen use time is associated with poorer mental well-being. This protocol describes a cross-sectional study of a pan-India, web-based convenience sample of young adults (18-24 y) with access to digital devices with a screen and a minimum of secondary school education. Participants will be recruited through people in the professional networks of the investigators, which includes pediatricians. The survey will also be distributed via the social media pages of our organization (X [X Corp], Instagram [Meta], Facebook [Meta], etc). Sociodemographic details will be collected through a questionnaire designed by the authors; screen use time and patterns will be assessed using an adaptation of the Screen Time Questionnaire to include data on different apps and websites used on digital devices; and mental health parameters will be gauged using the Warwick-Edinburgh Mental Well-Being Scale, Generalized Anxiety Disorder Scale, Perceived Stress Scale, and Patient Health Questionnaire. For statistical analysis, we will consider the following variables: (1) the primary independent variable is screen use time; (2) other independent variables include age, gender, residence: rural or urban, educational qualifications, employment status, stress associated with familial financial status, average sleep time, number of people living in a house or rooms in that house, BMI, substance use, and past psychiatric history; and (3) dependent variables include mental well-being, depression, anxiety, and perceived stress. To quantify the association between screen use time and mental health, we will perform a Bayesian multivariate multiple regression analysis that models the possibility of multiple alternative hypotheses while accounting for relevant sociodemographic covariables. The survey instrument has been designed, and feedback has been obtained from the domain experts and members of our organization whose profile is similar to the potential study participants. The final data received after this study has been conducted will be analyzed and shared. As of January 2023, we have not yet initiated the data collection. Based on the findings of this study, we will be able to establish a correlation between device- and use-specific screen use time and various mental health parameters. This will provide a direction to develop screen use time and mental health guidelines among young adults. PRR1-10.2196/39707.

Neurofibroma detection using joint imaging genomics and ensemble learning

Accurate detection of neurofibromas is crucial for quantitative monitoring of tumor progression and surgical assessment. This paper proposes a method for neurofibroma detection in whole-body magnetic resonance imaging (WBMRI) using joint imaging genomics and ensemble learning. Firstly, we enhance texture features through a combination of image sharpening, filtering, brightness adjustment, and contrast enhancement. Then, we employ a weighted boxes fusion (WBF) technique based on test-time augmentation (TTA) under a single model and further integrate multiple models using the dual fusion approach of TTA and WBF. For segmentation, we utilize minimum bounding boxes based on segmentation masks for position calibration. Finally, false positive tumor regions are further eliminated through imaging genomics features. The experimental MRI data is obtained from collaboration between Harvard Medical School and domestic tertiary hospitals, comprising 158 cases with a total of 1380 tumors. Five-fold cross-validation is conducted with segmentation annotations completed by domain experts. Compared to the best results of single models, our proposed method achieves a 10.1% increase in average precision (AP), 7.8% increase in sensitivity, reduction of average false positives to 3.58, a decrease of 17.68, and an 8.5% improvement in competitive performance metric (CPM). This method effectively enhances the accuracy of neurofibroma detection and is applicable to detecting tumors and lesions in other medical imaging applications.

Morphological evaluation of β-Ti-precipitation and its link to the mechanical properties of Ti–6Al–4V after laser powder bed fusion and subsequent heat treatments

The microstructural properties of additively manufactured Ti–6Al–4V depend strongly on the heat treatment after Laser Powder Bed Fusion. Among others the evaluation of morphological characteristics of β-Ti precipitates allows a fast and objective way to distinguish between different metallurgical conditions with different mechanical properties based on an automated analysis of high-resolution SEM images. High temperature heat treatment and hot isostatic pressing improve the mechanical properties while also reducing the scattering of results, leading to better predictable performance. Similar and subjectively not distinguishable microstructural conditions are observed after applying low temperature heat treatments of different sample batches with comparable UTS of app. 1100 MPa but very different elongation at break between 0.3 % and 6.5 %. Advanced classification methods based on machine learning and classical statistics allow a reliable distinction of almost identical microstructural conditions. Their specific characteristics in terms of required expert domain knowledge, general performance and interpretability of their results are discussed.

Probing perfection: The relentless art of meddling for pulmonary airway segmentation from HRCT via a human-AI collaboration based active learning method

In the realm of pulmonary tracheal segmentation, the scarcity of annotated data stands as a prevalent pain point in most medical segmentation endeavors. Concurrently, most Deep Learning (DL) methodologies employed in this domain invariably grapple with other dual challenges: the inherent opacity of ‘black box’ models and the ongoing pursuit of performance enhancement. In response to these intertwined challenges, the core concept of our Human-Computer Interaction (HCI) based learning models (RS_UNet, LC_UNet, UUNet and WD_UNet) hinge on the versatile combination of diverse query strategies and an array of deep learning models. We train four HCI models based on the initial training dataset and sequentially repeat the following steps 1–4: (1) Query Strategy: Our proposed HCI models selects those samples which contribute the most additional representative information when labeled in each iteration of the query strategy (showing the names and sequence numbers of the samples to be annotated). Additionally, in this phase, the model selects the unlabeled samples with the greatest predictive disparity by calculating the Wasserstein Distance, Least Confidence, Entropy Sampling, and Random Sampling. (2) Central line correction: The selected samples in previous stage are then used for domain expert correction of the system-generated tracheal central lines in each training round. (3) Update training dataset: When domain experts are involved in each epoch of the DL model's training iterations, they update the training dataset with greater precision after each epoch, thereby enhancing the trustworthiness of the ‘black box’ DL model and improving the performance of models. (4) Model training: Proposed HCI model is trained using the updated training dataset and an enhanced version of existing UNet.Experimental results validate the effectiveness of this Human-Computer Interaction-based approaches, demonstrating that our proposed WD-UNet, LC-UNet, UUNet, RS-UNet achieve comparable or even superior performance than the state-of-the-art DL models, such as WD-UNet with only 15 %–35 % of the training data, leading to substantial reductions (65 %–85 % reduction of annotation effort) in physician annotation time.

Decoding rail derailments: Unraveling the weighted factors influencing safety and sustainability using the best-worst method

The rail network is essential for sustainable transportation, offering various advantages such as reduced greenhouse gas emissions and congestion relief. However, ensuring safety within the rail network is crucial for its long-term viability and public acceptance. Derailment incidents have significant implications for safety, efficiency, and sustainability. This study employs the Best-Worst Method (BWM) to identify and weigh the parameters affecting derailment incidents. The research methodology involved conducting an extensive literature review to extract influential parameters, which were subsequently classified. Additionally, a rigorous data collection process was undertaken to ensure the reliability of the findings. The BWM was then applied, utilizing the expertise of five carefully selected domain experts who met specific selection criteria based on their experience and reputation in the field of railway safety. This expert panel provided valuable insights to determine the relative importance of the identified parameters. The calculated weights revealed the criticality of factors such as fractures in railway lines, illegal rail width, unauthorized locomotive speed, and defects in wagon wheels. Conversely, falling cargo train parts, improper load distribution, and subsidence of the railway line had relatively lesser influence. The results of this study offer valuable information for decision-makers and stakeholders in the rail industry, facilitating resource allocation and the implementation of targeted strategies to enhance rail safety.

Evolution-aware Constraint Derivation Approach for Software Remodularization

Existing software clustering techniques tend to ignore prior knowledge from domain experts, leading to results (suggested big-bang remodularization actions) that can not acceptable to developers. Incorporating domain experts knowledge or constraints during clustering ensures the obtained modularization aligns with developers’ perspectives, enhancing software quality. However, manual review by knowledgeable domain experts for constraint generation is time-consuming and labor-intensive. In this paper, we propose an evolution-aware constraint derivation approach, Escort, which automatically derives clustering constraints based on the evolutionary history from the analyzed software. Specifically, Escort can serve as an alternative approach to derive implicit and explicit constraints in situations where domain experts are absent. In the subsequent constrained clustering process, Escort can be considered as a framework to help supplement and enhance various unconstrained clustering techniques to improve their accuracy and reliability. We evaluate Escort based on both quantitative and qualitative analysis. In quantitative validation, Escort, using generated clustering constraints, outperforms seven classic unconstrained clustering techniques. Qualitatively, a survey with developers from five IT companies indicates that 89% agree with Escort’s clustering constraints. We also evaluate the utility of refactoring suggestions from our constrained clustering approach, with 54% acknowledged by project developers, either implemented or planned for future releases.

Surveillance of Health Care-Associated Violence Using Natural Language Processing.

Patient and family violent outbursts toward staff, caregivers, or through self-harm, have increased during the ongoing behavioral health crisis. These health care-associated violence (HAV) episodes are likely under-reported. We sought to assess the feasibility of using nursing notes to identify under-reported HAV episodes. We extracted nursing notes across inpatient units at 2 hospitals for 2019: a pediatric tertiary care center and a community-based hospital. We used a workflow for narrative data processing using a natural language processing (NLP) assisted manual review process performed by domain experts (a nurse and a physician). We trained the NLP models on the tertiary care center data and validated it on the community hospital data. Finally, we applied these surveillance methods to real-time data for 2022 to assess reporting completeness of new cases. We used 70 981 notes from the tertiary care center for model building and internal validation and 19 332 notes from the community hospital for external validation. The final community hospital model sensitivity was 96.8% (95% CI 90.6% to 100%) and a specificity of 47.1% (39.6% to 54.6%) compared with manual review. We identified 31 HAV episodes in July to December 2022, of which 26 were reportable in accordance with the hospital internal criteria. Only 7 of 26 cases were reported by employees using the self-reporting system, all of which were identified by our surveillance process. NLP-assisted review is a feasible method for surveillance of under-reported HAV episodes, with implementation and usability that can be achieved even at a low information technology-resourced hospital setting.

CLRiuS: Contrastive Learning for intrinsically unordered Steel Scrap

There has been remarkable progress in the field of Deep Learning and Computer Vision, but there is a lack of freely available labeled data, especially when it comes to data for specific industrial applications. However, large volumes of structured, semi-structured and unstructured data are generated in industrial environments, from which meaningful representations can be learned. The effort required for manual labeling is extremely high and can often only be carried out by domain experts. Self-supervised methods have proven their effectiveness in recent years in a wide variety of areas such as natural language processing or computer vision. In contrast to supervised methods, self-supervised techniques are rarely used in real industrial applications. In this paper, we present a self-supervised contrastive learning approach that outperforms existing supervised approaches on the used scrap dataset. We use different types of augmentations to extract the fine-grained structures that are typical for this type of images of intrinsically unordered items. This extracts a wider range of features and encodes more aspects of the input image. This approach makes it possible to learn characteristics from images that are common for applications in the industry, such as quality control. In addition, we show that this self-supervised learning approach can be successfully applied to scene-like images for classification.

Towards Model-Driven Explainable Artificial Intelligence: Function Identification with Grammatical Evolution

Machine learning is a well-matured discipline, and exploration of datasets can be performed in an efficient way, leading to accurate and operational prediction and decision models. On the other hand, most methods tend to produce black-box-type models, which can be considered a serious drawback. This is so, since in case of numerous practical applications, it is also required to justify, explain, and uncover the inner decision mechanism so that an in-depth understanding of the causal and functional dependencies becomes possible and some responsibility for the decision can be considered. This paper addresses the critical need for model-driven eXplainable Artificial Intelligence (XAI) by exploring the limitations inherent in existing explanatory mechanisms, such as LIME or SHAP, which rely solely on input data. This seems to be an intrinsic limitation and a conceptual error, as no expert domain knowledge can come into play, and no analytical models of the phenomena under investigation are created. In order to deal with this issue, this paper puts forward the idea of building open, white-box explanatory models. To do that, we propose employing grammatical evolution tools combined with expert domain knowledge. The results demonstrate that the developed models can effectively explain the structure and behavior of decision models in terms of components, connections, causality, and simple functional dependencies.

MEDI-NET : CADRE BASÉ SUR LE CLOUD POUR SYSTÈME DE RÉCUPÉRATION DE DONNÉES MÉDICALES UTILISANT LE DEEP LEARNING

Medical data retrieval is becoming increasingly crucial, aiding physicians and domain experts in more effectively accessing knowledge and information related to medicine and facilitating informed decision-making. The centralized architectures need help with scalability and real-time indexing, leading to extended retrieval times and decreased efficiency. To address these issues, a novel MEdical Data retrieval using Inception resNET (MEDI-NET) has been proposed to retrieve medical data efficiently. The proposed system introduces a deep learning network for a dynamic poly-indexing model and concurrent indexing, ensuring the real-time retrieval of the latest medical records. EMRBots and MIMIC-III are the two datasets used to compare the performance of the proposed MEDI-NET approach with existing MRCG, HCAC-EHR, and FedCBMIR method, which is implemented using Python. The effectiveness of the proposed MEDI-NET approach has been determined using evaluation metrics such as precision, accuracy, recall, F1-score, indexing time, and retrieval time. Comparative analysis with existing methods demonstrates superior precision, accuracy, recall, and F1 score for the proposed method. Additionally, the proposed system exhibits reduced indexing and retrieval times, showcasing its efficiency in handling large-scale medical data. The proposed MEDI-NET approach's accuracy is 0.60 %, 9.87 %, and 21.1 % higher than the existing MRCG, HCAC-EHR, and FedCBMIR techniques, respectively.

A unified approach to publish semantic annotations of agricultural documents as knowledge graphs

The research results presented in this paper were obtained as part of the D2KAB project (Data to Knowledge in Agriculture and Biodiversity) which aims to develop semantic web-based tools to describe and make agronomical data actionable and accessible following the FAIR principles. We focus on constructing domain-specific Knowledge Graphs (KGs) from textual data sources, using Natural Language Processing (NLP) techniques to extract and structure relevant entities. Our approach is based on the formalization of a semantic data model using common linked open vocabularies such as the Web Annotation Ontology (OA) and the Provenance Ontology (PROV). The model was developed by formulating motivating scenarios and competency questions from domain experts. This model has been used to construct three different KGs from three distinct corpora: PubMed scientific publications on wheat and rice genetics and phenotyping, and French agricultural alert bulletins. The named entities to be recognized include genes, phenotypes, traits, genetic markers, taxa and phenological stages normalized using semantic resources such as the Wheat Trait and Phenotype Ontology (WTO), the French Crop Usage (FCU) thesaurus and the Plant Phenological Description Ontology (PPDO). Named entities were extracted using different NLP approaches and tools. The relevance of the semantic model was validated by implementing experts questions as SPARQL queries to be answered on the constructed RDF knowledge graphs. Our work demonstrates how domain-specific vocabularies and systematic querying of KGs can reveal hidden interactions and support agronomists in navigating vast amounts of data. The resources and transformation pipelines developed are publicly available in Git repositories.

A Reference Architecture for Reverse Logistics in the High-Tech Industry

With the pressing need to shift towards sustainable sociotechnical systems, high-tech equipment manufacturers are facing an increasing demand for accountability in reverse logistics. Implementation of reverse logistics introduces complexities, particularly in handling the unpredictability of product returns, coordination of the network, and communication among different stakeholders along the reverse chain. To navigate these complexities and facilitate decision-making, the objective of this paper is to design a reference architecture tailored to the high-tech manufacturing sector. The proposed reference architecture encompasses a variety of reverse logistics activities, ranging from installed base management to recovery processes, to achieve supply chain circularity, and couples these witha data-driven decision-making processes to enhance efficiency and reliability throughout the reverse chain. The reference architecture is instantiated in the context of a real company case published in the literature and evaluated by domain experts and practitioners in the target industry. Evaluation results from questionnaire analysis indicate very positive responses towards our proposed solution in terms of completeness, flexibility, integration, simplicity, understandability, and usability. Overall, the reference architecture supports reverse logistics practitioners in integrated decision-making, ranging from organisational to operational levels.

Predicting the Liquid Steel End-Point Temperature during the Vacuum Tank Degassing Process Using Machine Learning Modeling

The present work focuses on predicting the steel melt temperature following the vacuum treatment step in a vacuum tank degasser (VTD). The primary objective is to establish a comprehensive methodology for developing and validating machine learning (ML) models within this context. Another objective is to evaluate the model by analyzing the alignment of the SHAP values with metallurgical domain expectations, thereby validating the model’s predictions from a metallurgical perspective. The proposed methodology employs a Random Forest model, incorporating a grid search with domain-informed variables grouped into batches, and a robust model-selection criterion that ensures optimal predictive performance, while keeping the model as simple and stable as possible. Furthermore, the Shapley Additive Explanations (SHAP) algorithm is employed to interpret the model’s predictions. The selected model achieved a mean adjusted R2 of 0.631 and a hit ratio of 75.3% for a prediction error within ±5 °C. Despite the moderate predictive performance, SHAP highlighted several aspects consistent with metallurgical domain expertise, emphasizing the importance of domain knowledge in interpreting ML models. Improving data quality and refining the model framework could enhance predictive performance.