In the early phases of a construction project, generating accurate and timely quotations is important for assessing feasibility. Delays or significant revisions in quotations can lead to project cancellations, resulting in lost business opportunities. To address this challenge, we propose a machine learning framework called Jointly Trained Automation of Explainable Construction Material Knowledge ( JACK ), developed in collaboration with a construction company to estimate material requirements. Our methodology begins with pre-processing estimation data, where construction materials are categorised into high-level types to facilitate more efficient learning. To support this process, open-source synthetic data generators were developed to help clarify structural patterns for JACK , which employs a cascaded learning approach during training. The evaluation phase leverages joint training to enhance model efficiency and presents results across 207 construction projects. We also investigate the effects of dropout layers, regression trees and synthetic data augmentation on prediction accuracy. Finally, we compare JACK against traditional regression-based methods using a separate project set, where it demonstrates competitive performance. Overall, JACK achieves low error rates across a range of material types, with performance gains largely attributed to the benefits of cascaded learning.

Clustering data in uncertain contexts is increasingly essential for addressing real-world challenges. This paper develops a novel incremental clustering approach, termed Incremental Possibilistic K-Modes with possibilistic membership degrees (IncPKM), designed to handle the arrival of new objects over time within a possibilistic framework. The proposed method utilizes possibility theory to manage uncertainty, allowing objects to belong to multiple clusters with varying degrees of possibility rather than fixed assignments. Unlike standard methods, IncPKM performs re-clustering as new data becomes available without re-initializing the entire process from scratch, significantly reducing computational overhead. Experimental evaluations using categorical datasets from the UCI Machine Learning Repository demonstrate that IncPKM maintains high accuracy while being more efficient than both the standard k-modes and nonincremental possibilistic k-modes methods.

Learning from imbalanced data remains a persistent challenge in machine learning, as standard classifiers tend to favor the majority class and may overlook minority-class patterns of primary interest. Existing solutions often rely on resampling or algorithmic modifications that alter the original data distribution or increase model complexity. Based on three-way Bayesian confirmation, this paper proposes conceptual models for imbalanced classification that preserve the original data while providing interpretable decision support. The proposed approach integrates the output of a standard binary classifier with confirmation-based reasoning derived from three-way Bayesian confirmation theory. Through different integration mechanisms, we develop conceptual models for strong and weak three-way classifiers. The strong model requires simultaneous confirmation of the predicted class and disconfirmation of the alternative class, whereas the weak model imposes milder confirmation constraints to improve coverage. Both models produce one of three outcomes: majority decision, minority decision, or non-commitment. Future work will focus on empirical evaluation and practical implementation of the proposed framework.

Digitization of medical records often relies on smartphone photographs of printed reports, producing images degraded by blur, shadows, and other noise. Conventional OCR systems, optimized for clean scans, perform poorly under such real-world conditions. This study evaluates compact multimodal language models as privacy-preserving alternatives for transcribing noisy clinical documents. Using obstetric ultrasound reports written in regionally inflected medical English common to Indian healthcare settings, we compare eight systems in terms of transcription accuracy, noise sensitivity, numeric accuracy, and computational efficiency. Compact multimodal models consistently outperform both classical and neural OCR pipelines. Despite higher computational costs, their robustness and linguistic adaptability position them as viable candidates for on-premises healthcare digitization.

BackgroundSudden cardiac arrest is a major cause of mortality, necessitating immediate and high-quality cardiopulmonary resuscitation (CPR) for improved survival rates. High-quality CPR is defined by chest compressions at a rate of 100-120 per minute and a depth of 50-60 mm. Monitoring and maintaining these parameters in real time during emergencies remain a challenge. ObjectiveThis study introduces a neural network model designed to predict and assess CPR quality using accelerometer data from a smartwatch. MethodsThe study involved 83 participants performing CPR on mannequins, with accelerometer data collected via smartwatches worn by the participants. These data were aligned with gold-standard data from the mannequins. The accelerometer-derived compression data were segmented into 5-second intervals for training the neural network models. A total of 1226 neural network models were developed, incorporating variations in hyperparameters and dataset configurations to optimize performance. ResultsThe optimal model demonstrated the capability to accurately predict the number of compressions and the average compression depth within a 5-second interval. The model achieved an accuracy of ±3.8 mm for compression depth and an average deviation of 0.8 compressions. The results indicated that the neural network model could accurately assess CPR quality metrics, surpassing other models discussed in the literature. The large and diverse dataset used in this study contributed to the robustness and reliability of the model. ConclusionsThis study validates the efficacy of a neural network model in accurately predicting CPR metrics using smartwatch accelerometer data. The model outperforms previous methods and shows promise for real-time feedback during CPR. Future work involves deploying the model directly on smartwatches for real-time application, potentially improving sudden cardiac arrest survival rates through immediate and accurate feedback on CPR quality.

Clustering short text embeddings is a foundational task in natural language processing, yet remains challenging due to the need to specify the number of clusters in advance. We introduce a scalable spectral method that estimates the number of clusters directly from the structure of the Laplacian eigenspectrum, constructed using cosine similarities and guided by an adaptive sampling strategy. This sampling approach enables our estimator to efficiently scale to large datasets without sacrificing reliability. To support intrinsic evaluation of cluster quality without ground-truth labels, we propose the Cohesion Ratio, a simple and interpretable evaluation metric that quantifies how much intra-cluster similarity exceeds the global similarity background. It has an information-theoretic motivation inspired by mutual information, and in our experiments it correlates closely with extrinsic measures such as normalized mutual information and homogeneity. Extensive experiments on six short-text datasets and four modern embedding models show that standard algorithms like K-Means and HAC, when guided by our estimator, significantly outperform popular parameter-light methods such as HDBSCAN, OPTICS, and Leiden. These results demonstrate the practical value of our spectral estimator and Cohesion Ratio for unsupervised organization and evaluation of short text data. Implementation of our estimator of k and Cohesion Ratio, along with code for reproducing the experiments, is available at https://anonymous.4open.science/r/towards_clustering-0C2E.

This study examines the performance of Large Language Models (LLMs) in Aspect-Based Sentiment Analysis (ABSA), with a focus on implicit aspect extraction in a novel domain. Using a synthetic sports feedback dataset, we evaluate open-weight LLMs' ability to extract aspect-polarity pairs and propose a metric to facilitate the evaluation of aspect extraction with generative models. Our findings highlight both the potential and limitations of LLMs in the ABSA task.

This paper explores the advancements and applications of language models in healthcare, focusing on their clinical use cases. It examines the evolution from early encoder-based systems requiring extensive fine-tuning to state-of-the-art large language and multimodal models capable of integrating text and visual data through in-context learning. The analysis emphasizes locally deployable models, which enhance data privacy and operational autonomy, and their applications in tasks such as text generation, classification, information extraction, and conversational systems. The paper also highlights a structured organization of tasks and a tiered ethical approach, providing a valuable resource for researchers and practitioners, while discussing key challenges related to ethics, evaluation, and implementation.

We present a comparative analysis of the parseability of structured outputs generated by small language models for open attribute-value extraction from clinical notes. We evaluate three widely used serialization formats: JSON, YAML, and XML, and find that JSON consistently yields the highest parseability. Structural robustness improves with targeted prompting and larger models, but declines for longer documents and certain note types. Our error analysis identifies recurring format-specific failure patterns. These findings offer practical guidance for selecting serialization formats and designing prompts when deploying language models in privacy-sensitive clinical settings.

In computer vision, most data are captured in 2D formats, limiting spatial understanding in real-world applications. This presents a challenge for fields such as architecture, construction, and robotics, where interpreting spatial relationships from minimal visual input is increasingly essential. This survey reviews recent advancements in extracting 3D features from 2D imagery, a critical task in these domains, where spatial accuracy and object orientation directly impact performance. We focus on three core areas: (1) disposition estimation, determining object pose; (2) joint modeling, constructing skeletal representations; and (3) scene reconstruction, generating spatially accurate environments. Each category is evaluated based on input modalities, performance metrics, and code availability. By providing a unified overview of these techniques, this paper highlights their practical value in enabling 3D reasoning from conventional 2D data.

In the construction industry, a project typically begins with the creation of two-dimensional (2D) building plans, defining the client’s specifications. Using these plans, a digital three-dimensional (3D) model is developed to visualize the anticipated outcome and to verify the model’s alignment with the client’s expectations. The process of converting from 2D to 3D can become time-intensive if there is a need for modifications or if the project’s overall complexity is high. To enhance efficiency and accuracy, this research introduces an end-to-end framework referred to as BIRD which stands for Building Image Reconstruction and Dimensioning. BIRD is capable of accepting five 2D perspective drawings of a building as inputs and generating a proportionate 3D model of the building envelope as an output. This is accomplished through the integration of multiple techniques that use convolutional neural networks to extract a refined set of line segments, identify measurements, and align each perspective with the floor plan drawing. The key contributions of this study includes: (1) a novel deep learning model designed for the identification of line segments in building plans; (2) novel algorithms that facilitate the generation of information required for 3D modeling; (3) an end-to-end framework for building reconstruction; and (4) novel performance metrics specifically tailored for the 2D to 3D conversion challenge. The practical application of this research was validated through the use of complete building plans provided by an industry partner. In summary, it was observed that BIRD demonstrated high accuracy in the creation of 3D visualizations, highlighting its real-world efficacy.

Due to the competitive nature of the construction industry, the efficiency of requirement analysis is important in enhancing client satisfaction and a company’s reputation. For example, determining the optimal configuration of panels (generally called panelization) that form the structure of a building is one aspect of cost estimation. However, existing methods typically rely on rule-based approaches that may lead to suboptimal material usage, particularly in complex designs featuring angled walls and openings. Such inefficiency can increase costs and environmental impact due to unnecessary material waste. To address these challenges, this research proposes a Panelization Algorithm for Architectural Designs, referred to as PAAD, which utilizes a genetic evolutionary strategy built on the 2D bin packing problem. This method is designed to balance between strict adherence to manufacturing constraints and the objective of optimizing material usage. PAAD starts with multiple potential solutions within the predefined problem space, facilitating dynamic exploration of panel configurations. It approaches structural rules as flexible constraints, making necessary corrections in post-processing, and through iterative developments, the algorithm refines panel sets to minimize material use. The methodology is validated through an analysis against an industry implementation and expert-derived solutions, highlighting PAAD’s ability to surpass existing results and reduce the need for manual corrections. Additionally, to motivate future research, a synthetic data generator, the architectural drawing encodings used, and a preliminary interface are also introduced. This not only highlights the algorithm’s practical applicability but also encourages its use in real-world scenarios.

In the event of cardiac arrest, providing immediate, high-quality cardiopulmonary resuscitation (CPR) and applying a defibrillator are crucial for patient care. High-quality CPR is defined by chest compressions at a rate of 100–120 per minute and a compression depth of 50–60 mm. However, during an emergency, monitoring the count and depth of compressions poses a significant challenge for individuals administering CPR. This study introduces a neural network model designed to predict and assess the quality of CPR utilizing accelerometer data from a participant’s smartwatch. This research involved collecting real-world chest compression data from 83 participants performing CPR on a mannequin, with accelerometer data captured via smartwatches worn by the participants. This data was employed to train the model against a gold-standard dataset from the mannequin. The accelerometer-derived compression data were aligned with those from the mannequin dataset. Subsequently, the data were segmented into five-second intervals to facilitate training the neural network models. Throughout the study, 1,226 neural network models were developed, incorporating variations in hyperparameters and the dataset. The optimal model demonstrated the capability to accurately predict the number of compressions and the average compression depth within a five-second interval, achieving an accuracy of ±3.8 mm and an average deviation in compression count of 0.8. The study validates the efficacy of a neural network model in accurately predicting CPR metrics, outperforming other models discussed in the literature and involving a considerably large participant base. The ethics application for this research received approval from TBRHSC (REB 2022519), allowing the collection and use of participant data for research purposes. Furthermore, all participants gave written consent for their data to be collected and used in this study.

The surgical services provided in the operating rooms of the hospital are an essential part of the healthcare system. These services are usually life-threatening and time-sensitive, and require highly trained surgeons and staff, as well as the latest medical equipment. Furthermore, because of the high cost of these resources, hospitals can have only a limited number of operating rooms and staff members. Thus, it is crucial to optimize various aspects of operating room functions to maximize overall utilization.This survey summarizes various optimization models proposed in the literature for such problems faced in operating rooms. The goal of this work is to provide researchers with a guide for further research in the field. Methods: This survey includes articles from Pubmed, since 2010. The search queries were related to the terms scheduling operating room, optimization model, and queuing. More than 400 articles were found, and the authors filtered the articles based on their relevance to this survey.The analysis found that a) the studies are usually very specific to optimizing a particular problem related to the operating room, b) datasets are not available in the literature and it is difficult to conduct comparative analysis, similarly c) the source code is not available on publicly available repositories like GitHub, and d) it is difficult to replicate the studies and establish benchmarks.

Background: Out of hospital cardiac arrest (OHCA) causes close to 400,000 deaths every year in North America, and it is also a leading cause of death among young athletes.OHCA is a treatable medical condition, and the patient's survival chances can be increased if immediate treatment is provided to the patient.However, non-treatment of the patient leads to a dramatic decline in survival chances at 10% per minute.Currently, various technologies are being used, and many more are being researched to reduce the time to provide early treatment to the patient.Objective: This survey focuses on summarizing various available technologies for use during OHCA.This survey focuses on evaluating technologies used in each step of the OHCA process.Methods: In this survey, articles were searched using the term "ohca" on Google Scholar and more than 18,000 articles were found.The articles were further filtered using keywords for each stage of the OHCA process (2,128).For each step, articles were filtered again using author developed method to select articles relevant to each technology for each step of the OHCA process (339).Finally, articles were manually filtered by authors based on their relevance and 112 articles were used in this survey.Results: The technologies that exist today work independently and are not linked with the other steps of the OHCA process.Integration between these technologies could help in reducing time and increase the survival chances of the patient.Also, if it is found that some of the proposed solutions are experimental in nature and not meant to be used in the real-world OHCA scenarios.

Background In a sudden cardiac arrest, starting CPR and applying an AED immediately are the two highest resuscitation priorities. Many existing mobile applications have been developed to assist users in locating a nearby AED. However, these applications do not provide indoor navigation to the AED location. The time required to locate an AED inside a building due to a lack of indoor navigation systems will reduce the patient’s chance of survival. The existing indoor navigation solutions either require special hardware, a large dataset or a significant amount of initial work. These requirements make these systems not viable for implementation on a large-scale. Methods The proposed system collects Wi-Fi information from the existing devices and the path’s magnetic information using a smartphone to guide the user from a starting point to an AED. The information collected is processed using four techniques: turn detection method, Magnetic data pattern matching method, Wi-Fi fingerprinting method and Closest Wi-Fi location method to estimate user location. The user location estimations from all four techniques are further processed to determine the user’s location on the path, which is then used to guide the user to the AED location. Results The four techniques used in the proposed system Turn detection , Magnetic data pattern matching , Closest Wi-Fi location and Wi-Fi fingerprinting can individually achieve the accuracy of 80% with the error distance ± 9.4 m, ± 2.4 m, ± 4.6 m, and ± 4.6 m respectively. These four techniques, applied individually, may not always provide stable results. Combining these techniques results in a robust system with an overall accuracy of 80% with an error distance of ± 2.74 m. In comparison, the proposed system’s accuracy is higher than the existing systems that use Wi-Fi and magnetic data. Conclusion This research proposes a novel approach that requires no special hardware, large scale data or significant initial work to provide indoor navigation. The proposed system AEDNav can achieve an accuracy similar to the existing indoor navigation systems. Implementing this indoor navigation system could reduce the time to locate an AED and ultimately increase patient survival during sudden cardiac arrest.

Background When an Out-of-Hospital Cardiac Arrest (OHCA) incident is reported to emergency services, the 911 agent dispatches Emergency Medical Services to the location and activates responder network system (RNS), if the option is available. The RNS notifies all the registered users in the vicinity of the cardiac arrest patient by sending alerts to their mobile devices, which contains the location of the emergency. The main objective of this research is to find the best match between the user who could support the OHCA patient. Methods For performing matching among the user and the AEDs, we used Bipartite Matching and Integer Linear Programming. However, these approaches take a longer processing time; therefore, a new method Preprocessed Integer Linear Programming is proposed that solves the problem faster than the other two techniques. Results The average processing time for the experimentation data was 1850 s using Bipartite matching, 32 s using the Integer Linear Programming and 2 s when using the Preprocessed Integer Linear Programming method. The proposed algorithm performs matching among users and AEDs faster than the existing matching algorithm and thus allowing it to be used in the real world. Conclusion: This research proposes an efficient algorithm that will allow matching of users with AED in real-time during cardiac emergency. Implementation of this system can help in reducing the time to resuscitate the patient.

User feedback is an important aspect of any social platform. The platform needs to adapt based on user preferences, eliminating the need for user input over time. This paper describes the experimentation with feedback received from users on their comfort level with temperatures in their work environment. The building energy management system will adapt itself so that users do not need to adjust the temperatures. This study addresses a number of issues with the user feedback modeling including managing a data imbalance with the help of fuzzy clustering, ability to transfer the user models between rooms and across different buildings, as well as, predicting a degree of discomfort. The experiments are based on user feedback collected from two commercial buildings.