Browse Items (16488 total)
Sort by:
-
Explainable Information Retrieval Techniques in Academic Search Engines
Due to the rapid increase in scholarly publications globally, researchers rely heavily on specialized academic search engines to gather pertinent information. However, the algorithms used in many of these systems create a black box that breaches transparency, significantly eroding user trust and interpretability. Thanks to XIR, Explainable Information Retrieval, this issue has become a thing of the past. Users can now receive easily understood rationales for why documents were retrieved and how the documents were ranked. This work examines the various XIR techniques integrated into academic search tools, assesses their application methods, and analyzes how effectively they enhance users' understanding, satisfaction, decision-making, and information processing. The paper also formulates a central proposal, which incorporates important elements of XIR, and highlights remnant problems that require deeper analysis. The Research Publication, www.trp.org.in. -
Explainable Hybrid Deep Learning Framework with Multimodal Inputs for Diabetic Retinopathy Detection
Diabetic Retinopathy (DR) is a leading cause of vision loss, making accurate and interpretable detection critical. This study proposes a hybrid interpretable machinedeep learning framework that integrates multimodal data for enhanced DR severity classification. The model combines unstructured fundus images from EyePACS, Messidor, and APTOS with structured clinical and lifestyle variables such as age, sex, HbA1c, BMI, blood pressure, and diabetes duration. Fundus images undergo preprocessing through resizing, normalization, augmentation, and noise reduction, while clinical data are imputed, normalized, and one-hot encoded. For feature extraction, EfficientNetV2, ResNet50, and Swin Transformer are applied to images, and XGBoost, LightGBM, and TabNet to clinical data. Features are fused via concatenation and attention, followed by classification using Logistic Regression, Random Forest, and MLP. Explainability is provided by Grad-CAM for imaging data and SHAP/LIME for clinical data, supporting clinical interpretability. The proposed model outperformed unimodal baselines, achieving 99.34% accuracy, 98.5% precision, 98.0% recall, 99.0% specificity, 98.2% F1-score, and 0.99 AUC-ROC, with a 10% gain over ResNet50 alone. Performance improvements included a 9% increase in recall and 8% in F1-score, alongside excellent calibration. Confusion matrix analysis confirmed balanced severity detection, and clinicians validated the interpretability outputs. This framework demonstrates robust accuracy, generalization, and clinical applicability for DR screening. 2026, An-Najah National University. All rights reserved. -
Explainable Artificial Intelligence: Frameworks for Ensuring the Trustworthiness
The growing computer power and ubiquity of big data are allowing Artificial Intelligence (AI) to gain widespread adoption and applicability in a wide range of sectors. The absence of an explanation for the conclusions made by today's AI algorithms is a significant disadvantage in crucial decision-making systems. For example, existing black-box AI systems are vulnerable to bias and adversarial assaults, which can taint the learning and inference processes. Explainable AI (XAI) is a recent trend in AI algorithms that gives explanations for their AI conclusions. Many contemporary AI systems have been shown to be vulnerable to undetectable assaults, biased against underrepresented groups, and deficient in user privacy protection. These flaws damage the user experience and undermine people's faith in all AI systems. This study proposes a systematic way to tie the social science notions of trust to the technology employed in AI-based services and products. 2024 IEEE. -
Explainable artificial intelligence in epilepsy management: Unveiling the model interpretability
The field of epileptic seizure classification has witnessed significant advancements in the use of electroencephalogram (EEG) data for accurate and timely diagnoses. This study introduces a comprehensive framework for EEG-based seizure classification, encompassing data preprocessing and the application of machine learning techniques, specifically the supervised learning classifier known as Extreme Gradient Boosting (Xgboost). Machine learning methods have shown promising accuracy in binary classification tasks, particularly in distinguishing between seizure and healthy EEG signals. However, the need for a robust explanation of these results and decision-making processes is imperative for technical verification and clinical validation, especially for potential clinical applications. Explainable Artificial Intelligence (XAI) emerges as a critical component in addressing this need. In this chapter, we propose and discuss a binary classification model that leverages Xgboost to classify EEG signals as either Seizure or normal, a crucial aspect in epilepsy diagnosis. XAI techniques such as LIME (Local Interpretable Model-agnostic Explanations) and SHAP (Shapley Additive Explanations) are incorporated to elucidate the model's predictions. LIME offers localized interpretability by creating surrogate models for individual predictions, revealing the essential EEG features influencing each classification decision. Conversely, SHAP provides a global perspective on feature importance, shedding light on the collective impact of EEG features on classification outcomes. The synergy between LIME and SHAP enhances our understanding of the model's predictions and the intricate nuances within EEG data. This research highlights the transformative potential of LIME and SHAP in EEG-based seizure classification. The integration of XAI techniques not only enhances the transparency and interpretability of the model but also empowers clinicians and researchers to make more informed decisions, ultimately improving patient care and outcomes in epilepsy management. By bridging the gap between complex EEG data and actionable insights, this study marks a significant paradigm shift in the application of XAI techniques in medical diagnostics. It paves the way for a new era in epilepsy diagnosis and management, where advanced machine learning models guided by LIME and SHAP play a crucial role in revolutionizing healthcare practices. 2025 Elsevier Inc. All rights reserved. -
Explainable artificial intelligence framework for wind turbine fault detection using random forest Extreme gradient boosting hybrid model
Though wind energy has great promise for clean energy generation in India, operational inefficiencies and underutilization still present major obstacles. Although installed wind capacity exceeds 51. 3 GW, actual power generation is still significantly lower than predicted mostly because of weak fault detection and maintenance techniques. Existing machine learning (ML) methods offer high accuracy but typically lack transparency in their forecasts, therefore making it difficult for engineers to correctly interpret and act on model outputs. This research aims to develop an understandable and high-performance anomaly detection model using real-time SCADA data from an Indian wind plant. This research aims to develop an understandable and high-performance anomaly detection model using real-time SCADA data from an Indian wind plant. A hybrid ensemble approach integrating Random Forest and XGBoost is proposed, combined with Local Interpretable Model-Agnostic Explanations (LIME) to provide local interpretability of predictions. The model was trained and evaluated on actual SCADA data using SelectKBest for feature selection, SMOTE for handling class imbalance, and RandomizedSearchCV for hyperparameter optimization. The tuned hybrid model achieved outstanding performance, with an accuracy of 0.9995, F1-score of 0.9995, and minimal error rates (MAE and MSE = 0.00052). LIME-based interpretability highlighted key features driving predictions, with Nacelle Temperature and Gearbox Temperature consistently emerging as critical indicators of turbine braking events, underscoring the importance of thermal variables in fault diagnosis. The findings suggest that interpretable machine learning not only enhances root cause analysis but also supports proactive maintenance, particularly by emphasizing improvements to cooling systems to reduce thermal failures. By providing transparent and reliable insights, the proposed solution enables wind farm operators to make informed, timely decisions, thereby improving turbine reliability and energy yield. The framework is practical, explainable, and well-suited for deployment in smart wind farms, aligning with the United Nations Sustainable Development Goals, including SDG 7 (Affordable and Clean Energy), SDG 9 (Industry, Innovation, and Infrastructure), and SDG 12 (Responsible Consumption and Production) 2025 The Author(s) -
Explainable artificial intelligence enhanced quantum-inspired spider monkey optimization for a constrained portfolio optimization proble
Optimizing portfolios has consistently posed significant challenges while being an extensively researched subject in finance and accounting. This process requires selecting and distributing appropriate assets in alignment with a set of specified objectives. This nonlinear constraint issue is not effectively solvable using traditional methods. This paper investigates the use of spider monkey optimization, ageist spider monkey optimization, and a newly proposed enhanced spider monkey optimization technique for portfolio optimization problems. The explainability of the spider monkey optimization has been improved without compromising the optimization results. It has been observed that the proposed technique marginally enhances the results of spider monkey optimization and can improve trust and risk management in the portfolio optimization problem. Furthermore, a quantum-inspired version of the proposed method is also implemented, and the results are compared using three benchmarked datasets from Dow Jones, BSE, and NASDAQ. Experimental results obtained using these benchmark datasets demonstrate that the newly introduced technique within the quantum-inspired framework marginally outperforms all other methods in the classical and quantum-inspired domains. The Author(s), under exclusive licence to Springer Nature Switzerland AG 2025. -
Explainable AI Method for Cyber bullying Detection
People of all ages and genders are using social media platforms to engage themselves in all sorts of activities. People create profiles on online social networks in order to communicate with one another in this virtual environment. Hundreds or thousands of friends and followers are split across many profiles. Along with the virtual communication in this social media life, cyber-crimes also creep in many distinguished forms to grab user's information and emotionally degrade them with harassment and arrogant behavior. A set of machine learning methods are proposed and used to detect such a bullying behavior. Along with the detection of such an act, the model should also provide the logical reasoning of the evidence extracted. The explain ability of the models classification will give us a view of the way towards portraying a suspect as a bullier. This paper illustrates a machine learning model that works on a twitter data set to suggest the tweets as category bullying or non-bullying. LIME a tool to predict the interpretability of the model is used to depict the performance of model and provides explainability. 2022 IEEE. -
Explainable AI in Healthcare: A Hybrid CNN-ViT Approach for Pneumonia Detection Using SHAP
The adoption of Artificial Intelligence (AI) in healthcare has improved diagnostic accuracy, particularly in medical imaging. However, the opaque nature of deep learning models raises concerns about interpretability in high-stakes applications such as pneumonia diagnosis. This study proposes an Explainable AI (XAI) framework that integrates a Hybrid CNN-ViT architecture with SHAP (SHapley Additive Explanations) for pneumonia detection from chest X-rays. Our approach achieves competitive diagnostic performance (94% accuracy) while enhancing transparency by highlighting clinically relevant features such as lobar consolidations and ground-glass opacities. By grounding explanations in established radiological findings, the framework supports clinical trust and regulatory compliance. This work contributes to bridging the gap between AI performance and medical accountability, positioning explainable deep learning as a trustworthy tool for real-world healthcare deployment. The Author(s), under exclusive license to Springer Nature Switzerland AG 2026. -
Explainable AI for Secure and Trustworthy Autonomous Network Management
Rise of AI-driven autonomous networks for managing complex, dynamic infrastructures. While AI optimizes performance, it acts as a black box. This lack of transparency undermines trust and security, making it challenging to validate decisions, detect adversarial attacks, and understand why an AI model made a specific routing, security, or resource allocation decision. Security blind spots face significant challenges in detecting subtle adversarial manipulations or policy exploits because the reasoning behind the model's decisions is hidden. Additionally, poor diagnosability occurs when a network fault or performance degradation occurs, making root cause analysis slow and complex. Hence, the network operators are hesitant to cede control to systems whose actions they cannot verify or audit. Explainable AI (XAI) is critical for bridging this gap, ensuring management decisions are transparent, interpretable, and defensible. The proposed model makes real-Time management decisions. This model uses post-hoc techniques to generate explanations for each decision. It presents actionable insights and cross-references explanations against security policies and known threat patterns to flag anomalous reasoning. 2025 IEEE. -
Explainable AI for Heart Disease prediction: A Clinical Transparency Route Experiment
In this paper, a proposeable explainable machine learning procedure on estimating the danger of heart attack will be proposed with a stacked ensemble of XGBoost, Random Forest, and Multi-layered perceptron (MLP). The data set of UCI Heart Disease was preprocessed by normalization, imputation, and SMOTE to address the imbalance problem and the variables were optimized with the help of the feature engineering. The model performance was measured using accuracy, precision, recall, F1-score and ROC-AUC. In order to make the results more interpretable, Explainable AI were applied with SHAP and LIME, and the most relevant risk factors including troponin, cholesterol, and blood pressure were indicated.. In this paper, it is shown that ensemble learning in XAI can yield plausible, interpretable, and clinically practical data to complement enhanced cardiovascular diagnostics. 2025 IEEE. -
Explainable AI for Heart Disease prediction: A Clinical Transparency Route Experiment
In this paper, a proposeable explainable machine learning procedure on estimating the danger of heart attack will be proposed with a stacked ensemble of XGBoost, Random Forest, and Multi-layered perceptron (MLP). The data set of UCI Heart Disease was preprocessed by normalization, imputation, and SMOTE to address the imbalance problem and the variables were optimized with the help of the feature engineering. The model performance was measured using accuracy, precision, recall, F1-score and ROC-AUC. In order to make the results more interpretable, Explainable AI were applied with SHAP and LIME, and the most relevant risk factors including troponin, cholesterol, and blood pressure were indicated.. In this paper, it is shown that ensemble learning in XAI can yield plausible, interpretable, and clinically practical data to complement enhanced cardiovascular diagnostics. 2025 IEEE. -
Explainable AI for Diabetic Retinopathy Screening: Enhancing Clinician Trust in Deep Learning Predictions
Diabetic retinopathy (DR) remains a leading cause of preventable blindness worldwide, with early detection being critical for effective intervention. While deep learning models have demonstrated exceptional performance in automated DR screening, their black box nature has limited clinical adoption due to concerns about interpretability and trust. This paper presents a comprehensive explainable AI (XAI) framework that integrates multiple visualization techniques, including Gradient-weighted Class Activation Mapping (Grad-CAM), attention mechanisms, and feature attribution methods, to provide clinically meaningful explanations for DR predictions. We evaluate our approach on the publicly available EyePACS and Messidor-2 datasets, achieving 94.3% accuracy while generating interpretable heatmaps that highlight lesion-specific regions. A clinical validation study involving 15 ophthalmologists demonstrates that our XAI-augmented system increases diagnostic confidence by 23% and reduces review time by 31% compared to non-explainable models. Our findings suggest that transparent AI systems can effectively bridge the gap between algorithmic performance and clinical trust, paving the way for broader adoption of AI-assisted DR screening in healthcare settings. 2026 IEEE. -
Explainable AI and computational intelligence in healthcare: Application to clinical decision support and personalized medicine
Human intelligence system simulation has made significant strides in several areas, including clinical decision-making using medical imaging and electronic health records, health referral systems, discovering recommended medications and vaccines, recognizing prescribed errors, and real-time data analysis. Therefore it is essential to discover patterns and transfer knowledge in the medical domain. The obstacles at the level of data collection, data analysis, model development, decision-making, and ethical concerns need to be addressed. It is vital that clinical interpretation tools associated with both hardware and software employed by medical professionals be precisely examined when rendering decisions regarding diagnoses and therapies related to the diagnosis. Computer scientists generally lack training in medical concepts specific to their field. Another crucial aspect is that black box algorithms based on artificial and computational intelligence are opaque and devoid of logical justification. Owing to these limitations, the technique of eXplainable Artificial Intelligence (XAI) models is explored in this chapter, primarily focusing on improving the interpretability of computational models. Specific objectives of this chapter are to: a) discuss the role that CI techniques and methods in the construction of an intelligent health prediction system; b) demonstrate the multiple CI paradigms utilized in medical prediction; and c) present recent case studies to showcase the performance of the computational intelligent models. 2026 Elsevier Inc. All rights reserved. -
Experimenting with scalability of floodlight controller in software defined networks
Software Defined Network is the booming area of research in the domain of networking. With growing number of devices connecting to the global village of internet, it becomes inevitable to adapt to any new technology before testing its scalability in presence of dynamic circumstances. While a lot of research is going on to provide solution to overcome the limitations of the traditional network, it gives a call to research community to test the applicability and caliber to withstand the fault tolerance of the provided solution in the form of SDN Controllers. Out of existing multiple controllers providing the SDN functionalities to the network, one of the stellar controllers is Floodlight Controller. This paper is a contribution towards performance evaluation of scalability of the Floodlight Controller by implementing multiple scenarios experimented on the simulation tool of Mininet, Floodlight Controller and iPerf. Floodlight Controller is tested in the simulation environment by observing throughput and latency parameters of the controller and checked its performance in dynamic networking conditions over Mesh topology by exponentially increasing the number of nodes. 2017 IEEE. -
Experimenting with scalability of Beacon controller in software defined network
In traditional network, a developer cannot develop software programs to control the behavior of the network switches due to closed vendor specific configuration scripts. In order to bring out innovations and to make the switches programmable a new network architecture must be developed. This led to a new concept of Software Defined Networking(SDN). In Software defined networking architecture, the control plane is detached from the data plane of a switch. The controller is implemented using the control plane which takes the heavy lift of all the requests of the network. Few of the controllers used in SDN are Floodlight, Ryu, Beacon, Open Daylight etc. In this paper, authors are evaluating the performance of Beacon controller using scalability parameter on network emulation tool Mininet and IPERF. The experiments are performed on multiple scenarios of topology size range from 50 to 1000 nodes and further analyzing the controller performance. BEIESP. -
Experimenting with resilience and scalability of wifi mininet on small to large SDN networks
Today everything is getting digitized where people want to be wireless by all aspects. There is a high demand of WiFi in every sector. Highest influence on network planning of newly developed network infrastructure is of SDN to meet the futuristic needs of upcoming technology. As a result, newly developed networks have become more adaptive to dynamic circumstances along with enhanced flexibility. Being globally connected, it is inevitable to obtain adequate services from data centers through Wi-Fi support on SDN Networks, which is still a dream. Thus, the target of the experiment performed and presented by the authors of this paper is to implement WiFi support on SDN. Further, authors have also demonstrated the scalability and resilience of SDN based WiFi Network on Mininet by testing performance parameters in various dynamic scenarios. This paper will have a high impact on the end users as SDN technology can be implemented as last mile technology using WiFi SDN. BEIESP. -
Experimenting with ONOS scalability on software defined network
In traditional network, a developer cannot change the configuration of a router with software programs to control the behavior of the network switches due to closed vendor specific configuration scripts. In order to make the routers/switches programmable, a new architecture of network has to be developed and this gave rise to Software defined networks. It is the new architecture for Computer Networks in which, the old traditional architecture is slowly depreciated. It is very difficult to adapt new technology especially to decide upon which controller has to be considered and what may be its scalability to compete the dynamic circumstances of networks. Many researches are working on possible solutions and look upon SDN to overcome the traditional network limitations. There are many SDN controllers existing amongst them, some are OpenDaylight, Floodlight, Onos, Ryu, Beacon etc. From the existing multiple controllers serving the SDN services to the network, Onos is one of the Controller. ONOS can be deployed on Docker container and it is accessed using its IP as a host. In this paper, authors are contributing for the evaluation of the Performance to check the Scalability of ONOS controller by taking many scenarios which are experimented on the simulation tool of Mininet, Onos Controller, Docker and iPerf. ONOS Controller?s simulated environments are observed for its throughput evaluated in dynamic conditions of a network over Mesh topology by gradually increasing the number of hosts until its supported by the system with optimum resource utilization. 2018, Institute of Advanced Scientific Research, Inc.. All rights reserved. -
Experimentally optimizing a spinning disk by manipulating its mass distribution and radius
The scientific method enables the experimental study of complex phenomena by isolating key variables. This work explores the significant properties of spinning bodies. Optimizing spinning disks is the primary aim of this work. Optimization is achieved by manipulating the moment of inertia (MOI) of the disk, allowing a longer duration of spin and lowering the rate of energy dissipation. Experiments are designed and conducted to explore the relationship between the radius and mass distribution of the disk and the angular deceleration experienced by it. Effects of the same on energy retention is analyzed. Empirical data is interpreted graphically while accounting for systematic and random uncertainties. Percentage change in duration of spin as a result of percentage change in physical quantities is studied. Moving mass away from the central axis of the spinning disk increases its duration of spin from a constant initial angular velocity. Energy retention is also improved. Increasing the radius of the disk increases the duration of spin and reduces the rate of energy dissipation. The above conclusions are drawn from experiments where the mass and thickness of the disk are controlled along with other necessary factors that can influence the results. The experiments confirm the existing theory relating to the moment of inertia, angular quantities, resistive torques and kinetic energy of spinning disks. The experiments provide insights into the behavior of spinning disks in practical situations, especially in problems concerned with optimization in the field of mechanical engineering. 2026 Veeresha et al., published by Paradigm. -
Experimental, FEA, and machine learning studies on wear behavior of LM13 aluminum hybrid composites reinforced with zircon and graphite
This paper examines applied load and zircon reinforcement influence on LM13 alloy composites wear behavior. LM13 was reinforced with 3?wt.% graphite with 3, 6, 9, and 12 weight percent of zircon utilizing a stir casting technique with a chill end to achieve unidirectional solidification. Wear tests were conducted on specimen's chill end using a pin-on-disc apparatus under loads of 30?N, to 70?N in steps of 10?N incremental. The results indicated that when the amount of zircon went up, the wear rate dropped, reaching a minimum at 9?wt.% zircon, then slightly increasing at 12?wt.%. Specifically, wear rate reduced from 4.2?10?3mm/Nm at 3?wt.% zircon to 2.7?10?3mm/Nm at 9?wt.% zircon, before rising to 3.5?10?3mm/Nm at 12?wt.%, establishing 9?wt.% zircon as the optimum reinforcement. Finite Element Analysis (FEA) had been used to simulate wear behavior, and its predictions aligned well with experimental data, with deviations under 5%. Both experimental and FEA results confirmed that wear rate increases proportionally with applied load. Additionally, machine learning techniques were employed to validate the observed trends, enhancing the reliability of the findings. Microstructural analysis through Field Emission Scanning Electron Microscopy showed evidence of plastic deformation and delamination at higher stress levels, compromising material integrity. Notably, the composite with 9?wt.% zircon exhibited reduced wear deformation and minimal microstructural damage, confirming its effectiveness in improving wear resistance. IMechE 2025 -
Experimental Verification of Gain and Bandwidth Enhancement of Fractal Contoured Metamaterial Inspired Antenna
The performance of any antenna cannot be completely assessed purely on the basis of simulation results. All simulations are made by assuming an ideal environment where the fabrication tolerances and practical losses are not accounted for. Therefore, evidencing the performance experimentally becomes a crucial step. In this work, the experimental validation of a fractal contoured square microstrip antenna with four ring metamaterial structure, hereon referred to as optimized metamaterial inspired square fractal antenna has been presented. It is an extension to previously designed antenna and aims to experimentally verify the enhanced gain and bandwidth of this antenna. The design and simulation of the proposed antenna was accomplished by using Ansys HFSS v18.2. The end-to-end antenna spread area is 23 mm x 23 mm on a 46 mm x 28 mm x 1.6 mm FR4 substrate with ?r of 4.4. The simulated design was fabricated using Nvis 72 Prototyping Machine and measured in an anechoic chamber facility using vector network analyzer. The antenna resonates with the deepest S11 of-39.5 dB in a broad bandwidth of 2.53 GHz from 2.265 GHz to 4.79 GHz with experimental verification. The proposed antenna provides an enhanced gain of 8.81 dB at the most popularly used frequency of 2.5 GHz. The simulation and experimental results of resonance, gain and radiation pattern are found to agree maximally. The fractional bandwidth offered by this proposed antenna is 72.28%. The experimental validation confirms enhanced gain-bandwidth performance in a wide resonance band. Hence, this antenna is well recommended for wireless, energy harvesting rectenna and sub-6 GHz (2.5 GHz to 4.20 GHz) 5G applications. 2022, Advanced Electromagnetics. All rights reserved.