Explainable Deep Learning Research Articles

DeePathNet: a transformer-based deep learning model integrating multi-omic data with cancer pathways.

Multi-omic data analysis incorporating machine learning has the potential to significantly improve cancer diagnosis and prognosis. Traditional machine learning methods are usually limited to omic measurements, omitting existing domain knowledge, such as the biological networks that link molecular entities in various omic data types. Here we develop a Transformer-based explainable deep learning model, DeePathNet, which integrates cancer-specific pathway information into multi-omic data analysis. Using a variety of big datasets, including ProCan-DepMapSanger, CCLE, and TCGA, we demonstrate and validate that DeePathNet outperforms traditional methods for predicting drug response and classifying cancer type and subtype. Combining biomedical knowledge and state-of-the-art deep learning methods, DeePathNet enables biomarker discovery at the pathway level, maximizing the power of data-driven approaches to cancer research. DeePathNet is available on GitHub at https://github.com/CMRI-ProCan/DeePathNet.

An explainable deep learning model to predict partial anomalous pulmonary venous connection for patients with atrial septal defect

BackgroundPatients with partial anomalous pulmonary venous connection (PAPVC) usually present asymptomatic and accompanied by intricate anatomical types, which results in missed diagnosis from atrial septal defect (ASD). The present study aimed to explore the predictive variables of PAPVC from patients with ASD and constructed an explainable prediction model based on deep learning.MethodsThe retrospective study included 834 inpatients with ASD in Women and Children's Hospital, Qingdao University from January 2018 to January 2023. They were separated into two groups based on the presence of PAPVC. Propensity score matching and SMOTE were used to balance the baseline data between groups. The differential variables between the two groups were determined by univariate logistic regression. The patients were randomly divided into the training set and the validation set in a ratio of 8:2. Support vector machines (SVM), Random forest, Decision tree, XGBoost, and LightGBM were used to build models by differential variables. The classification performance of models was compared. Split, gain and SHAP were used to measure the importance of differential variables and improve the interpretability of the model. Moreover, a portion of the patients was included in the validation set to test the performance of the selected models.ResultsThree hundred twenty-eight patients with ASD and patients with 82 PAPVC were included in the training set and the validation set, respectively. The selection of 10 differential variables was based on univariate logistic regression, including right atrial diameter (longitudinal axis and transverse axis), right ventricular diameter, left atrial diameter, left ventricular end-diastolic diameter, left ventricular end-systolic diameter, P-wave voltage, P-wave interval PR interval, and QRS-wave voltage. In the classification model established based on differential variables, the LightGBM model achieved the highest performance on the validation set (AUC = 0.93). Based on variables importance analysis, the LightGBM-Clinic model was retrained by P-wave voltage, P-wave interval, PR interval, QRS wave interval, and right ventricular diameter, and performed excellently (AUC = 0.90). The AUC of the LightGBM-Clinic model was 0.87 in the test set.ConclusionIn this study, the LightGBM model performs excellently in determining whether patients with ASD are accompanied by PAPVC. ECG parameters such as P-wave voltage were important to predictive value and enhance the explainability of the model.

A genetic programming Rician noise reduction and explainable deep learning model for Alzheimer's diseases severity prediction.

Degradation of magnetic resonance imaging (MRI) remains a challenging issue, with noise being a key damaging component introduced due to a variety of environmental and mechanical factors. The aim of this research work is to addresses the issue of noise reduction and to predict Alzheimer's disease detection efficiently. First, we present a genetic programming (GP) technique for reducing Rician noise in MRI images to pre-process the dataset. To effectively reduce Rician noise, this GP approach combines a Feature Extraction component, GP Optimal Expression, and an Optimum Removal Estimation component. In the second phase, we design and develop an explainable Deep Learning framework. This framework uses a local data-driven interpretation technique based on SHAP values to investigate the relationship between the neural network's estimated AD diagnosis and the input MRI images. In addition, we handle class distribution by combining an oversampling strategy with a minority approach. Several assessment metrics are used to analyze the performance of our proposed model. The proposed method is tested on a variety of medical samples, and the results are compared to those obtained using other comparable approaches. We also test and compare our model to three cutting-edge models: DenseNet169, VGGNet15, and Inceptionv3. The empirical results show that our proposed model outperforms others, particularly in handling basic structures with limited spectral features, lower computational complexity, and less overfitting. This research worked addressed Rician noise issue in MRI images and predict AD severity prediction using explainable deep learning framework.

Explainable deep learning on multi-target time series forecasting: an air pollution use case

Urban air pollution represents a significant threat to public health and the environment, with nitrogen oxides, ozone, and particulate matter being among the most harmful pollutants. These contribute to respiratory and cardiovascular diseases, particularly in urban areas with high traffic and elevated temperatures. Machine learning, especially deep learning, shows promise in enhancing the prediction accuracy of prediction of pollutant's concentrations. However, the “black box” nature of these models often limits their interpretability, which is crucial for informed decision-making. Our study introduces a Temporal Selection Layer technique within deep learning models for time series forecasting to tackle this issue. This technique not only improves prediction accuracy by embedding feature selection directly into the neural network, but also enhances interpretability and reduces computational costs. In particular, we applied this method to hourly concentration data of pollutants, including particulate matter, ozone, and nitrogen oxides, from five urban monitoring sites in Graz, Austria. These concentrations were used as target variables to predict, while identifying the most relevant features and periods that affect prediction accuracy. Comparative analysis with other embedded feature selection methods showed that the Temporal Selection Layer significantly enhances both model effectiveness and transparency. Additionally, we applied explainable techniques to evaluate the impact of weather and time-related factors on air pollution, which also helped assess feature importance. The results show that our approach improves both prediction accuracy and model interpretability, leading finally to more effective pollution management strategies.

An eXplainable Deep Learning Model for Multi-Modal MRI Grading of IDH-Mutant Astrocytomas

IDH-Mutant-astrocytomas are malignant brain glioma tumors. They are graded as lower (grades II and III) or higher (grade IV) according to their rate of growth and molecular features. Conventional IDH grading requires pathology examination, which is invasive, costly, and time-consuming. Non-invasive MRI methods are widely used but require a high level of interpretation for tumor grading. Recent advances in deep learning have shown potential to aid and improve the accuracy and efficiency of tumor classification. This study proposes a novel non-invasive approach for IDH-Mutant-astrocytomas grading using Light-weight Attention Network (LAN) that integrates multi-modal MRI data, including both conventional and advanced MR modalities. The study addresses inter-modality heterogeneity using Principal Component Analysis (PCA) while minimizing computational complexity. LAN uses a 3D Convolutional Neural Network (CNN) and a volumetric attention mechanism to extract and classify local and global tumor patterns. This is followed by an eXplainable AI approach that use SHapley Additive exPlanations (SHAP) values to understand the model decisions. LAN model outperforms other pre-trained state-of-the-art models, with an overall accuracy of 0.84. The advanced modalities Arterial Spin Labeling (ASL) and the Diffusion-weighted imaging(DWI) SHAP score contributes significantly in improving tumor heterogeneity visualization for aggressive grade II along with conventional MRIs. This work demonstrates the feasibility of integrating explainable deep learning with multi-modal MRI data for precise and comprehensible early glioma grading, potentially leading to better clinical decision-making.

An explainable deep learning approach for stock market trend prediction

Given the intricate nature of stock forecasting as well as the inherent risks and uncertainties, analysis of market trends is necessary to capitalize on optimal investment opportunities for profit maximization and timely disinvestment for loss minimization. In this work, we propose a deep learning model for predicting five distinct stock market trends: upward, downward, double top, rounded bottom, and rounded top. The proposed model surpasses common benchmarks, including support vector machine, random forest, and logistic regression, achieving an average accuracy of 94.9%, compared to 85.7% for random forest, 60.07% for support vector machine, and 52.45% for logistic regression. Furthermore the proposed model excels in F1-score, with a 94.85% performance, compared to 77.95% for random forest, 21.02% for support vector machine and 46.23% for logistic regression, across four real world diverse datasets. Additionally, we employ explainable AI (XAI) techniques, SHAP and LIME, to enhance interpretability, enabling stakeholders to understand the key factors driving predictions. The SHAP analysis reveals the top 10 most important/influential features, enabling feature reduction while maintaining performance. Interestingly, while accuracy slightly decreases with top 10 features, precision, recall, and F1-score improve, suggesting a trade-off between comprehensiveness and performance. These results demonstrate the potential for practical application in financial decision-making, providing a balance between interpretability and predictive power that can support investors in risk management and strategic planning.

Explainable Deep Learning for Breast Cancer Classification and Localisation

Explainable Deep Learning for Breast Cancer Classification and Localisation

Identification of lineage-specific cis-trans regulatory networks related to kiwifruit ripening initiation.

Previous research on the ripening process of many fruit crop varieties typically involved analyses of the conserved genetic factors among species. However, even for seemingly identical ripening processes, the associated gene expression networks often evolved independently, as reflected by the diversity in the interactions between transcription factors (TFs) and the targeted cis-regulatory elements (CREs). In this study, explainable deep learning (DL) frameworks were used to predict expression patterns on the basis of CREs in promoter sequences. We initially screened potential lineage-specific CRE-TF interactions influencing the kiwifruit ripening process, which is triggered by ethylene, similar to the corresponding processes in other climacteric fruit crops. Some novel regulatory relationships affecting ethylene-induced fruit ripening were identified. Specifically, ABI5-like bZIP, G2-like, and MYB81-like TFs were revealed as trans-factors modulating the expression of representative ethylene signaling/biosynthesis-related genes (e.g., ACS1, ERT2, and ERF143). Transient reporter assays and DNA affinity purification sequencing (DAP-Seq) analyses validated these CRE-TF interactions and their regulatory relationships. A comparative analysis with co-expression networking suggested that this DL-based screening can identify regulatory networks independently of co-expression patterns. Our results highlight the utility of an explainable DL approach for identifying novel CRE-TF interactions. These imply that fruit crop species may have evolved lineage-specific fruit ripening-related cis-trans regulatory networks.

AI driven interpretable deep learning based fetal health classification

In this study, a deep learning model is proposed for the classification of fetal health into 3 categories: Normal, suspect, and pathological. The primary objective is to utilize the power of deep learning to improve the efficiency and effectiveness of diagnostic processes. A deep neural network (DNN) model is proposed for fetal health analysis using data obtained from Cardiotocography (CTG). A dataset containing 21 attributes is used to carry out this work. The model incorporates multiple hidden layers, augmented with batch normalization and dropout layers for improved generalization. This study assesses the model's interpretation ability in fetal health classification using explainable deep learning. This enhances transparency in decision-making of the classifier model by leveraging feature importance and feature saliency analysis, fostering trust and facilitating the clinical adoption of fetal health assessments. Our proposed model demonstrates a remarkable performance with 0.99 accuracy, 0.93 sensitivity, 0.93 specificity, 0.96 AUC, 0.93 precision, and 0.93 F1 scores in classifying fetal health. We also performed comparative analysis with six other models including Logistic Regression, KNN, SVM, Naive Bayes, Random Forest, and Gradient Boosting to assess and compare the effectiveness of our model and the accuracies of 0.89, 0.88, 0.90, 081, 0.93, and 0.93 were achieved respectively by these baseline models. The results revealed that our proposed model outperformed all the baseline models in terms of accuracy. This indicates the potential of deep learning in improving fetal health assessment and contributing to the field of obstetrics by providing a robust tool for early risk detection.

Empowering Brain Tumor Diagnosis through Explainable Deep Learning

Brain tumors are among the most lethal diseases, and early detection is crucial for improving patient outcomes. Currently, magnetic resonance imaging (MRI) is the most effective method for early brain tumor detection due to its superior imaging quality for soft tissues. However, manual analysis of brain MRI scans is prone to errors, largely influenced by the radiologists’ experience and fatigue. To address these challenges, computer-aided diagnosis (CAD) systems are more significant. These advanced computer vision techniques such as deep learning provide accurate predictions based on medical images, enhancing diagnostic precision and reliability. This paper presents a novel CAD framework for multi-class brain tumor classification. The framework employs six pre-trained deep learning models as the base and incorporates comprehensive data preprocessing and augmentation strategies to enhance computational efficiency. To address issues related to transparency and interpretability in deep learning models, Gradient-weighted Class Activation Mapping (Grad-CAM) is utilized to visualize the decision-making processes involved in tumor classification from MRI scans. Additionally, a user-friendly Brain Tumor Detection System has been developed using Streamlit, demonstrating its practical applicability in real-world settings and providing a valuable tool for clinicians. All simulation results are derived from a public benchmark dataset, showing that the proposed framework achieves state-of-the-art performance, with accuracy approaching 99% in ResNet-50, Xception, and InceptionV3 models.

Explainable deep learning for glaucoma detection: Efficient dense net with quantum-based Dwarf Mongoose optimization

Glaucoma, a progressive eye disease, leads to irreversible vision loss due to optic nerve damage. This study aims to develop a more explainable and interpretable deep learning-based model for the detection and classification of glaucoma, addressing the current limitations in the field. The proposed approach utilizes a modified DenseNet-201 architecture, incorporating an additional transition layer for enhanced glaucoma classification. A pre-trained Efficient DenseNet is applied, and a reweighted cross-entropy loss function is used to handle the class imbalance in the training data. Quantum-based Dwarf Mongoose Optimization (QDMOA) is employed to fine-tune the model’s parameters, minimizing over fitting and improving generalization on small datasets. Experimental results demonstrate that the proposed method effectively detects and classifies two types of glaucoma. The use of dense connections with regularization significantly reduces over fitting, and the reweighted loss function improves model performance on imbalanced data. The developed deep learning model offers improved accuracy in glaucoma detection and classification while addressing the need for more interpretable and reliable models in medical applications. This research provides a practical tool for automated glaucoma screening, which could support ophthalmologists in early diagnosis and treatment, reducing the risk of irreversible vision loss.

Leveraging explainable deep learning methodologies to elucidate the biological underpinnings of Huntington’s disease using single-cell RNA sequencing data

BackgroundHuntington’s disease (HD) is a hereditary neurological disorder caused by mutations in HTT, leading to neuronal degeneration. Traditionally, HD is associated with the misfolding and aggregation of mutant huntingtin due to an extended polyglutamine domain encoded by an expanded CAG tract. However, recent research has also highlighted the role of global transcriptional dysregulation in HD pathology. However, understanding the intricate relationship between mRNA expression and HD at the cellular level remains challenging. Our study aimed to elucidate the underlying mechanisms of HD pathology using single-cell sequencing data.ResultsWe used single-cell RNA sequencing analysis to determine differential gene expression patterns between healthy and HD cells. HD cells were effectively modeled using a residual neural network (ResNet), which outperformed traditional and convolutional neural networks. Despite the efficacy of our approach, the F1 score for the test set was 96.53%. Using the SHapley Additive exPlanations (SHAP) algorithm, we identified genes influencing HD prediction and revealed their roles in HD pathobiology, such as in the regulation of cellular iron metabolism and mitochondrial function. SHAP analysis also revealed low-abundance genes that were overlooked by traditional differential expression analysis, emphasizing its effectiveness in identifying biologically relevant genes for distinguishing between healthy and HD cells. Overall, the integration of single-cell RNA sequencing data and deep learning models provides valuable insights into HD pathology.ConclusionWe developed the model capable of analyzing HD at single-cell transcriptomic level.

Automated Detection of Central Retinal Artery Occlusion Using OCT Imaging via Explainable Deep Learning

ObjectiveTo demonstrate the capability of a deep learning model to detect central retinal artery occlusion (CRAO), a retinal pathology with significant clinical urgency, using optical coherence tomography (OCT) data. DesignRetrospective, external validation study analyzing OCT and clinical baseline data of two institutions via deep learning classification analysis. SubjectsPatients presenting to the University Medical Center Tübingen and the University Medical Center Hamburg-Eppendorf in Germany. MethodsOCT-data of patients suffering from CRAO, differential diagnosis with (sub-)acute visual loss (central retinal vein occlusion, diabetic macular edema, non-arteritic ischemic optic neuropathy) and from controls were expertly graded and distinguished into the three groups.Our methodological approach involved a nested multiclass 5-fold cross-validation classification scheme. Main Outcome MeasuresArea under the curve (AUC) ResultsThe optimal performance of our algorithm was observed using 30 epochs, complemented by an early stopping mechanism to prevent overfitting. Our model followed a multiclass approach, distinguishing among the three different classes: Control CRAO and differential diagnoses. The evaluation was conducted by the “one vs. all” area under the receiver operating characteristics (ROC) curve (AUC) method. The results demonstrated AUC of 0.96 (95% confidence interval (CI): ± 0.01), 0.99 (95% CI: ± 0.00), 0.90 (95% CI: ± 0.03) for each class, respectively. ConclusionsOur machine learning algorithm (MLA) exhibited a high AUC, as well as sensitivity and specificity in detecting CRAO and the differential classes, respectively. These findings underscore the potential for deploying MLAs in the identification of less common etiologies within an acute emergency clinical setting.

An Explainable Deep Learning Approach for Classifying Monkeypox Disease by Leveraging Skin Lesion Image Data

According to the World Health Organization's (WHO) external situation report on the multi-country outbreak of Monkeypox in 2023, from 11 countries in Southeast Asia Regions, Thailand recorded the highest reported cases, totaling 461. The ongoing Monkeypox outbreak has raised significant public health concerns due to its rapid spread across several nations. Early detection and diagnosis are imperative for effectively treating and controlling Monkeypox. Given this context, this study aimed to determine the most efficient model for detecting Monkeypox by employing interpretable deep learning techniques. This study utilizes deep learning techniques to diagnose Monkeypox based on images of skin lesions. We evaluate based on four models—convolutional neural network (CNN), gated recurrent unit (GRU), long short-term memory (LSTM), and bidirectional long short term memory (BiLSTM)—using a publicly available dataset. Additionally, we incorporate Local Interpretable Model-Agnostic Explanations (LIME) and techniques for explainable AI, facilitating visual interpretation of model predictions for healthcare practitioners. The CNN model's performance and LSTM model's performance have an accuracy of 100%, while the GRU model's performance and BiLSTM model's performance have an accuracy of 99.88% and 99.45%. Our findings demonstrate the effectiveness of deep learning models, including the suggested CNN model leveraging the pre-trained MobileNetV2 and LSTM. These models can play a pivotal role in combating the Monkeypox virus. Doi: 10.28991/ESJ-2024-08-05-013 Full Text: PDF

Enhancing transparency in data-driven urban pluvial flood prediction using an explainable CNN model

Mitigating severe losses caused by pluvial floods in urban areas with dense population and property requires effective flood prediction for emergency measures. Physics-based models face issues with low computational efficiency for real‐time flood prediction. An alternative approach for rapid prediction instead of physics-based models is to predict from a data-driven perspective. However, data-driven approaches for urban flood prediction are often perceived as “black box” and might raise concerns. In this study, we propose an explainable deep learning (DL) approach for rapid urban pluvial flood prediction with enhanced transparency using a convolutional neural network (CNN) and the explainable artificial intelligence (AI) framework Shapley additive explanation (SHAP). We process a systematic stepwise feature selection process and establish a CNN model considering topography, drainage networks and rainfall to predict maximum inundation depths. Then, SHAP is applied to provide trustworthy explanations for the decision making in model results. The results show that: 1) Forward selection can offer insights into selecting effective input variables for improved predictions and promote understanding of DL modelling. The spatial pattern of inundation depths predicted by the proposed CNN model shows good agreement with those predicted by the physics-based model, demonstrated by average correlation coefficient (CC) and mean absolute error (MAE) values of 0.982 and 0.021 m, respectively. 2) The CNN model substantially outperforms the physics-based model in computational speed when using the same hardware, achieving speedups of 210 times on GPU and 51 times on CPU in the case study (575167 grid cells, 14.38 km2). Particularly, it can still achieve good performance on a CPU-only standard laptop without high-performance hardware, with only a modest increase in computational time. 3) The SHAP explainable analysis confirms that the CNN model correctly captures the relationships between input variables and water depth, revealing a reasonable decision-making process, enhancing its transparency. The explainable DL approach incorporating SHAP for rapid urban pluvial flood prediction is promising to build trust among stakeholders and provide a trustworthy reference for prompt measures aiming at saving lives and protecting assets during flood emergencies. Additionally, the proposed DL approach can potentially be further expanded to analyze the causes of urban flooding events and serve as a foundation for exploring the transferability of data-driven urban flood prediction, providing benefits for better urban flood risk management.

An explainable deep learning model for automated classification and localization of microrobots by functionality using ultrasound images

The rapid advancements of untethered microrobots offer exciting opportunities in fields such as targeted drug delivery and minimally invasive surgical procedures. However, several challenges remain, especially in achieving precise localization and classification of microrobots within living organisms using ultrasound (US) imaging. Current US-based detection algorithms often suffer from inaccurate visual feedback, causing positioning errors. This paper presents a novel explainable deep learning model for the localization and classification of eight different types of microrobots using US images. We introduce the Attention-Fused Bottleneck Module (AFBM), which enhances feature extraction and improves the performance of microrobot classification and localization tasks. Our model consistently outperforms baseline models such as YOLOR, YOLOv5-C3HB, YOLOv5-TBH, YOLOv5 m, and YOLOv7. The proposed model achieved mean Average Precision (mAP) of 0.861 and 0.909 at an IoU threshold of 0.95 which is 2% and 1.5% higher than the YOLOv5 m model in training and testing, respectively. Multi-thresh IoU analysis was performed at IoU thresholds of 0.6, 0.75, and 0.95, and demonstrated that the microrobot localization accuracy of our model is superior. A robustness analysis was performed based on high and low frequencies, gain, and speckle in our test data set, and our model demonstrated higher overall accuracy. UsingScore-CAM in our framework enhances interpretability, allowing for transparent insights into the model’s decision-making process. Our work signifies a notable advancement in microrobot classification and detection, with potential applications in real-world scenarios using the newly available USMicroMagset dataset for benchmarking.

In-field disease symptom detection and localisation using explainable deep learning: Use case for downy mildew in grapevine

Diseases and pests in agriculture significantly impact crop yield and quality. Downy mildew (Plasmopara viticola) is a particular noteworthy example in grapevines. Traditional detection methods are laborious, subjective and time-consuming. Consequently, a technological solution based on artificial intelligence, would provide higher levels of reproducibility and sampling. The aim of this work was to develop an interpretable, automated method for detection and localisation of plant disease symptoms under field conditions. Images of the grapevine canopy were taken in 14 commercial vineyard plots under a range of lightning conditions, including both static and on-the-go settings. The images were processed using a sliding window, classifying sub-images into areas with and without downy mildew. Transfer learning, fine-tuning and data augmentation were employed to automate the classification, comparing convolutional neural networks (CNNs) and vision transformers (ViT). Subsequently, the trained model was integrated into the sliding window to localise regions within the canopy images exhibiting symptoms of downy mildew. Model predictions were interpreted using explainable artificial intelligence (XAI) methods. The EfficientNetV2S model achieved an accuracy of 91 % and an F1-score of 0.92 when classifying image areas and an Intersection over Union (IoU) of 0.83 when locating symptomatic areas. This method showed promising results, enabling automatic and explainable detection and localisation of plant diseases in complex conditions. The straightforward labelling process facilitated adaptation to new conditions, making it suitable for different crops and diseases. Integration into mobile platforms could enhance disease management and reduce the spread of pathogens, making a significant advance in agricultural technology.

Explainable Deep Learning Approach for Mpox Skin Lesion Detection with Grad-CAM

Mpox is a viral zoonotic disease that presents with skin lesions similar to other conditions like chickenpox, measles, and hand-foot-mouth disease, making accurate diagnosis challenging. Early and precise detection of mpox is critical for effective treatment and outbreak control, particularly in resource-limited settings where traditional diagnostic methods are often unavailable. While deep learning models have been applied successfully in medical imaging, their use in mpox detection remains underexplored. To address this gap, we developed a deep learning-based approach using the ResNet50v2 model to classify mpox lesions alongside five other skin conditions. We also incorporated Grad-CAM (Gradient-weighted Class Activation Mapping) to enhance model interpretability. The results show that the ResNet50v2 model achieved an accuracy of 99.33%, precision of 99.34%, sensitivity of 99.33%, and an F1-score of 99.32% on a dataset of 1,594 images. Grad-CAM visualizations confirmed that the model focused on relevant lesion areas for its predictions. While the model performed exceptionally well overall, it struggled with misclassifications between visually similar diseases, such as chickenpox and mpox. These results demonstrate that AI-based diagnostic tools can provide reliable, interpretable support for clinicians, particularly in settings with limited access to specialized diagnostics. However, future work should focus on expanding datasets and improving the model's capacity to distinguish between similar conditions.

Explainable deep learning approach for advanced persistent threats (APTs) detection in cybersecurity: a review

In recent years, Advanced Persistent Threat (APT) attacks on network systems have increased through sophisticated fraud tactics. Traditional Intrusion Detection Systems (IDSs) suffer from low detection accuracy, high false-positive rates, and difficulty identifying unknown attacks such as remote-to-local (R2L) and user-to-root (U2R) attacks. This paper addresses these challenges by providing a foundational discussion of APTs and the limitations of existing detection methods. It then pivots to explore the novel integration of deep learning techniques and Explainable Artificial Intelligence (XAI) to improve APT detection. This paper aims to fill the gaps in the current research by providing a thorough analysis of how XAI methods, such as Shapley Additive Explanations (SHAP) and Local Interpretable Model-agnostic Explanations (LIME), can make black-box models more transparent and interpretable. The objective is to demonstrate the necessity of explainability in APT detection and propose solutions that enhance the trustworthiness and effectiveness of these models. It offers a critical analysis of existing approaches, highlights their strengths and limitations, and identifies open issues that require further research. This paper also suggests future research directions to combat evolving threats, paving the way for more effective and reliable cybersecurity solutions. Overall, this paper emphasizes the importance of explainability in enhancing the performance and trustworthiness of cybersecurity systems.

Explainable Deep Learning Approach for High Impedance Fault Localization in Resonant Distribution Networks Considering Quantization Noise

ABSTRACTIn addressing the quantization noise challenge in high impedance fault (HIF) localization within resonant distribution networks, we propose a cutting‐edge, explainable deep learning approach that significantly advances existing methods. This approach utilizes differential zero‐sequence voltage (DZSV) and zero‐sequence current (ZSC) and introduces a novel “Vague” classification to improve localization accuracy by effectively managing quantization noise‐distorted signals. This approach extends beyond the conventional binary classification of “Fault” and “Sound,” incorporating a multi‐scale feature attention (MFA) mechanism for enriched internal explainability and applying gradient‐weighted class activation mapping (Grad‐CAM) to visualize critical input areas precisely. Our model, validated in an industrial prototype, exhibits unparalleled adaptability across various environmental conditions, including environmental noise, variable sampling rates, and triggering deviations. Comparative analysis reveals that our approach outperforms existing methods in managing diverse fault scenarios.