Explainable Matrix - Visualization for Global and Local Interpretability of Random Forest Classification Ensembles

An Interpretable Adaptive Multiscale Attention Deep Neural Network for Tabular Data

Deep learning (DL) has been demonstrated to be a valuable tool for analyzing signals such as sounds and images, thanks to its capabilities of automatically extracting relevant patterns as well as its end-to-end training properties. When applied to tabular structured data, DL has exhibited some performance limitations compared to shallow learning techniques. This work presents a novel technique for tabular data called adaptive multiscale attention deep neural network architecture (also named excited attention). By exploiting parallel multilevel feature weighting, the adaptive multiscale attention can successfully learn the feature attention and thus achieve high levels of F1-score on seven different classification tasks (on small, medium, large, and very large datasets) and low mean absolute errors on four regression tasks of different size. In addition, adaptive multiscale attention provides four levels of explainability (i.e., comprehension of its learning process and therefore of its outcomes): 1) calculates attention weights to determine which layers are most important for given classes; 2) shows each feature's attention across all instances; 3) understands learned feature attention for each class to explore feature attention and behavior for specific classes; and 4) finds nonlinear correlations between co-behaving features to reduce dataset dimensionality and improve interpretability. These interpretability levels, in turn, allow for employing adaptive multiscale attention as a useful tool for feature ranking and feature selection.

AdversaFlow: Visual Red Teaming for Large Language Models with Multi-Level Adversarial Flow

Large Language Models (LLMs) are powerful but also raise significant security concerns, particularly regarding the harm they can cause, such as generating fake news that manipulates public opinion on social media and providing responses to unethical activities. Traditional red teaming approaches for identifying AI vulnerabilities rely on manual prompt construction and expertise. This paper introduces AdversaFlow, a novel visual analytics system designed to enhance LLM security against adversarial attacks through human-AI collaboration. AdversaFlow involves adversarial training between a target model and a red model, featuring unique multi-level adversarial flow and fluctuation path visualizations. These features provide insights into adversarial dynamics and LLM robustness, enabling experts to identify and mitigate vulnerabilities effectively. We present quantitative evaluations and case studies validating our system's utility and offering insights for future AI security solutions. Our method can enhance LLM security, supporting downstream scenarios like social media regulation by enabling more effective detection, monitoring, and mitigation of harmful content and behaviors.

Analysis of Macronutrients in Soil Using Impedimetric Multisensor Arrays

The need to increase food production to address the world population growth can only be fulfilled with precision agriculture strategies to increase crop yield with minimal expansion of the cultivated area. One example is site-specific fertilization based on accurate monitoring of soil nutrient levels, which can be made more cost-effective using sensors. This study developed an impedimetric multisensor array using ion-selective membranes to analyze soil samples enriched with macronutrients (N, P, and K), which is compared with another array based on layer-by-layer films. The results obtained from both devices are analyzed with multidimensional projection techniques and machine learning methods, where a decision tree model algorithm chooses the calibrations (best frequencies and sensors). The multicalibration space method indicates that both devices effectively distinguished all soil samples tested, with the ion-selective membrane setup presenting a higher sensitivity to K content. These findings pave the way for more environmentally friendly and efficient agricultural practices, facilitating the mapping of cropping areas for precise fertilizer application and optimized crop yield.

Multidimensional calibration spaces in Staphylococcus Aureus detection using chitosan-based genosensors and electronic tongue

Mastitis diagnosis can be made by detecting Staphylococcus aureus (S. aureus), which requires high sensitivity and selectivity. Here, we report on microfluidic genosensors and electronic tongues to detect S. aureus DNA using impedance spectroscopy with data analysis employing visual analytics and machine learning techniques. The genosensors were made with layer-by-layer films containing either 10 bilayers of chitosan/chondroitin sulfate or 8 bilayers of chitosan/sericin functionalized with an active layer of cpDNA S. aureus. The specific interactions leading to hybridization in these genosensors allowed for a low limit of detection of 5.90 × 10-19 mol/L. The electronic tongue had four sensing units made with 6-bilayer chitosan/chondroitin sulfate films, 10-bilayer chitosan/chondroitin sulfate, 8-bilayer chitosan/sericin, and 8-bilayer chitosan/gold nanoparticles modified with sericin. Despite the absence of specific interactions, various concentrations of DNA S. aureus could be distinguished when the impedance data were plotted using a dimensionality reduction technique. Selectivity of S. aureus DNA was confirmed using multidimensional calibration spaces, based on machine learning, with accuracy up to 89 % for the genosensors and 66 % for the electronic tongue. Hence, with these computational methods one may opt for the more expensive genosensors or the simpler and cheaper electronic tongue, depending on the sensitivity level required to diagnose mastitis.

Synergy of shaped-induced enhanced Raman scattering to improve surface-enhanced Raman scattering signal in the thiram molecule detection

Herein, we explore the combined effect of Shaped-Induced Enhanced Raman Scattering (SIERS) and Surface-Enhanced Raman Scattering (SERS) for detecting thiram molecules. We fabricated V-shaped microchannels on a silicon (100) substrate through a standard lithography and etching process. The analysis of SIERS@SERS was performed for Si-V substrates modified with AuNRs with different thiram concentrations, 10-7 to 10-10 mol/L. The spectra were collected for different regions of the Si-V substrates, i.e., in the inside, edge, between (flat top), and far from Si-V (coffee-ring AuNRs aggregation) to assess the performance of Si-V microchannels obtained. The IDMAP statistical projection reveals a higher silhouette coefficient of 0.91 for the inside of Si-V, indicating a more excellent spectral reproducibility with closer relative intensities. The device platform used in this study stands out as a robust option for commercial sensors, demonstrating exceptional sensitivity in detecting a diverse range of molecules, even at low concentrations.

Challenges involved in the application of artificial intelligence in gastroenterology: The race is on!

Gastroenterology is a particularly data-rich field, generating vast repositories of data that are a fruitful ground for artificial intelligence (AI) and machine learning (ML) applications. In this opinion review, we initially elaborate on the current status of the application of AI/ML-based software in gastroenterology. Currently, AI/ML-based models have been developed in the following applications: Models integrated into the clinical setting following real-time patient data flagging patients at high risk for developing a gastrointestinal disease, models employing non-invasive parameters that provide accurate diagnoses aiming to either replace, minimize, or refine the indications of endoscopy, models utilizing genomic data to diagnose various gastrointestinal diseases, computer-aided diagnosis systems facilitating the interpretation of endoscopy images, models to facilitate treatment allocation and predict the response to treatment, and finally, models in prognosis predicting complications, recurrence following treatment, and overall survival. Then, we elaborate on several challenges and how they may negatively impact the widespread application of AI in healthcare and gastroenterology. Specifically, we elaborate on concerns regarding accuracy, cost-effectiveness, cybersecurity, interpretability, oversight, and liability. While AI is unlikely to replace physicians, it will transform the skillset demanded by future physicians to practice. Thus, physicians are expected to engage with AI to avoid becoming obsolete.

Nanoarchitectonic E-Tongue of Electrospun Zein/Curcumin Carbon Dots for Detecting Staphylococcus aureusin Milk

We report a nanoarchitectonic electronic tongue made with flexible electrodes coated with curcumin carbon dots and zein electrospun nanofibers, which could detect Staphylococcus aureus(S. aureus) in milk using electrical impedance spectroscopy. Electronic tongues are based on the global selectivity concept in which the electrical responses of distinct sensing units are combined to provide a unique pattern, which in this case allowed the detection of S. aureus through non-specific interactions. The electronic tongue used here comprised 3 sensors with electrodes coated with zein nanofibers, carbon dots, and carbon dots with zein nanofibers. The capacitance data obtained with the three sensors were processed with a multidimensional projection technique referred to as interactive document mapping (IDMAP) and analyzed using the machine learning-based concept of multidimensional calibration space (MCS). The concentration of S. aureus could be determined with the sensing units, especially with the one containing zein as the limit of detection was 0.83 CFU/mL (CFU stands for colony-forming unit). This high sensitivity is attributed to molecular-level interactions between the protein zein and C-H groups in S. aureus according to polarization-modulated infrared reflection-absorption spectroscopy (PM-IRRAS) data. Using machine learning and IDMAP, we demonstrated the selectivity of the electronic tongue in distinguishing milk samples from mastitis-infected cows from milk collected from healthy cows, and from milk spiked with possible interferents. Calibration of the electronic tongue can also be reached with the MCS concept employing decision tree algorithms, with an 80.1% accuracy in the diagnosis of mastitis. The low-cost electronic tongue presented here may be exploited in diagnosing mastitis at early stages, with tests performed in the farms without requiring specialized laboratories or personnel.

Detecting dengue fever in children using online Rasch analysis to develop algorithms for parents: An APP development and usability study

BACKGROUND

Dengue fever (DF) is a significant public health concern in Asia. However, detecting the disease using traditional dichotomous criteria (i.e., absent vs present) can be extremely difficult. Convolutional neural networks (CNNs) and artificial neural networks (ANNs), due to their use of a large number of parameters for modeling, have shown the potential to improve prediction accuracy (ACC). To date, there has been no research conducted to understand item features and responses using online Rasch analysis. To verify the hypothesis that a combination of CNN, ANN, K-nearest-neighbor algorithm (KNN), and logistic regression (LR) can improve the ACC of DF prediction for children, further research is required.

METHODS

We extracted 19 feature variables related to DF symptoms from 177 pediatric patients, of whom 69 were diagnosed with DF. Using the RaschOnline technique for Rasch analysis, we examined 11 variables for their statistical significance in predicting the risk of DF. Based on 2 sets of data, 1 for training (80%) and the other for testing (20%), we calculated the prediction ACC by comparing the areas under the receiver operating characteristic curve (AUCs) between DF + and DF- in both sets. In the training set, we compared 2 scenarios: the combined scheme and individual algorithms.

RESULTS

Our findings indicate that visual displays of DF data are easily interpreted using Rasch analysis; the k-nearest neighbors algorithm has a lower AUC (<0.50); LR has a relatively higher AUC (0.70); all 3 algorithms have an almost equal AUC (=0.68), which is smaller than the individual algorithms of Naive Bayes, LR in raw data, and Naive Bayes in normalized data; and we developed an app to assist parents in detecting DF in children during the dengue season.

CONCLUSION

The development of an LR-based APP for the detection of DF in children has been completed. To help patients, family members, and clinicians differentiate DF from other febrile illnesses at an early stage, an 11-item model is proposed for developing the APP.

A hybrid anomaly detection method for high dimensional data

Anomaly detection of high-dimensional data is a challenge because the sparsity of the data distribution caused by high dimensionality hardly provides rich information distinguishing anomalous instances from normal instances. To address this, this article proposes an anomaly detection method combining an autoencoder and a sparse weighted least squares-support vector machine. First, the autoencoder is used to extract those low-dimensional features of high-dimensional data, thus reducing the dimension and the complexity of the searching space. Then, in the low-dimensional feature space obtained by the autoencoder, the sparse weighted least squares-support vector machine separates anomalous and normal features. Finally, the learned class labels to be used to distinguish normal instances and abnormal instances are outputed, thus achieving anomaly detection of high-dimensional data. The experiment results on real high-dimensional datasets show that the proposed method wins over competing methods in terms of anomaly detection ability. For high-dimensional data, using deep methods can reconstruct the layered feature space, which is beneficial for gaining those advanced anomaly detection results.

Colorimetric Detection of SARS-CoV-2 Using Plasmonic Biosensors and Smartphones

Low-cost, instrument-free colorimetric tests were developed to detect SARS-CoV-2 using plasmonic biosensors with Au nanoparticles functionalized with polyclonal antibodies (f-AuNPs). Intense color changes were noted with the naked eye owing to plasmon coupling when f-AuNPs form clusters on the virus, with high sensitivity and a detection limit of 0.28 PFU mL-1 (PFU stands for plaque-forming units) in human saliva. Plasmon coupling was corroborated with computer simulations using the finite-difference time-domain (FDTD) method. The strategies based on preparing plasmonic biosensors with f-AuNPs are robust to permit SARS-CoV-2 detection via dynamic light scattering and UV-vis spectroscopy without interference from other viruses, such as influenza and dengue viruses. The diagnosis was made with a smartphone app after processing the images collected from the smartphone camera, measuring the concentration of SARS-CoV-2. Both image processing and machine learning algorithms were found to provide COVID-19 diagnosis with 100% accuracy for saliva samples. In subsidiary experiments, we observed that the biosensor could be used to detect the virus in river waters without pretreatment. With fast responses and requiring small sample amounts (only 20 μL), these colorimetric tests can be deployed in any location within the point-of-care diagnosis paradigm for epidemiological control.

Visual Exploration of Machine Learning Model Behavior with Hierarchical Surrogate Rule Sets

One of the potential solutions for model interpretation is to train a surrogate model: a more transparent model that approximates the behavior of the model to be explained. Typically, classification rules or decision trees are used due to their logic-based expressions. However, decision trees can grow too deep, and rule sets can become too large to approximate a complex model. Unlike paths on a decision tree that must share ancestor nodes (conditions), rules are more flexible. However, the unstructured visual representation of rules makes it hard to make inferences across rules. In this paper, we focus on tabular data and present novel algorithmic and interactive solutions to address these issues. First, we present Hierarchical Surrogate Rules (HSR), an algorithm that generates hierarchical rules based on user-defined parameters. We also contribute SuRE, a visual analytics (VA) system that integrates HSR and an interactive surrogate rule visualization, the Feature-Aligned Tree, which depicts rules as trees while aligning features for easier comparison. We evaluate the algorithm in terms of parameter sensitivity, time performance, and comparison with surrogate decision trees and find that it scales reasonably well and overcomes the shortcomings of surrogate decision trees. We evaluate the visualization and the system through a usability study and an observational study with domain experts. Our investigation shows that the participants can use feature-aligned trees to perform non-trivial tasks with very high accuracy. We also discuss many interesting findings, including a rule analysis task characterization, that can be used for visualization design and future research.

Predicting Heavy Metal Concentrations in Shallow Aquifer Systems Based on Low-Cost Physiochemical Parameters Using Machine Learning Techniques

Monitoring ex-situ water parameters, namely heavy metals, needs time and laboratory work for water sampling and analytical processes, which can retard the response to ongoing pollution events. Previous studies have successfully applied fast modeling techniques such as artificial intelligence algorithms to predict heavy metals. However, neither low-cost feature predictability nor explainability assessments have been considered in the modeling process. This study proposes a reliable and explainable framework to find an effective model and feature set to predict heavy metals in groundwater. The integrated assessment framework has four steps: model selection uncertainty, feature selection uncertainty, predictive uncertainty, and model interpretability. The results show that Random Forest is the most suitable model, and quick-measure parameters can be used as predictors for arsenic (As), iron (Fe), and manganese (Mn). Although the model performance is auspicious, it likely produces significant uncertainties. The findings also demonstrate that arsenic is related to nutrients and spatial distribution, while Fe and Mn are affected by spatial distribution and salinity. Some limitations and suggestions are also discussed to improve the prediction accuracy and interpretability.

A Comparative Study of Two Rule-Based Explanation Methods for Diabetic Retinopathy Risk Assessment

Understanding the reasons behind the decisions of complex intelligent systems is crucial in many domains, especially in healthcare. Local explanation models analyse a decision on a single instance, by using the responses of the system to the points in its neighbourhood to build a surrogate model. This work makes a comparative analysis of the local explanations provided by two rule-based explanation methods on RETIPROGRAM, a system based on a fuzzy random forest that analyses the health record of a diabetic person to assess his/her degree of risk of developing diabetic retinopathy. The analysed explanation methods are C-LORE-F (a variant of LORE that builds a decision tree) and DRSA (a method based on rough sets that builds a set of rules). The explored methods gave good results in several metrics, although there is room for improvement in the generation of counterfactual examples.

Uncertainty-Based Rejection in Machine Learning: Implications for Model Development and Interpretability

Uncertainty is present in every single prediction of Machine Learning (ML) models. Uncertainty Quantification (UQ) is arguably relevant, in particular for safety-critical applications. Prior research focused on the development of methods to quantify uncertainty; however, less attention has been given to how to leverage the knowledge of uncertainty in the process of model development. This work focused on applying UQ into practice, closing the gap of its utility in the ML pipeline and giving insights into how UQ is used to improve model development and its interpretability. We identified three main research questions: (1) How can UQ contribute to choosing the most suitable model for a given classification task? (2) Can UQ be used to combine different models in a principled manner? (3) Can visualization techniques improve UQ’s interpretability? These questions are answered by applying several methods to quantify uncertainty in both a simulated dataset and a real-world dataset of Human Activity Recognition (HAR). Our results showed that uncertainty quantification can increase model robustness and interpretability.

A novel explainable COVID-19 diagnosis method by integration of feature selection with random forest

Several Artificial Intelligence-based models have been developed for COVID-19 disease diagnosis. In spite of the promise of artificial intelligence, there are very few models which bridge the gap between traditional human-centered diagnosis and the potential future of machine-centered disease diagnosis. Under the concept of human-computer interaction design, this study proposes a new explainable artificial intelligence method that exploits graph analysis for feature visualization and optimization for the purpose of COVID-19 diagnosis from blood test samples. In this developed model, an explainable decision forest classifier is employed to COVID-19 classification based on routinely available patient blood test data. The approach enables the clinician to use the decision tree and feature visualization to guide the explainability and interpretability of the prediction model. By utilizing this novel feature selection phase, the proposed diagnosis model will not only improve diagnosis accuracy but decrease the execution time as well.

Evaluating the Quality of Machine Learning Explanations: A Survey on Methods and Metrics

The most successful Machine Learning (ML) systems remain complex black boxes to end-users, and even experts are often unable to understand the rationale behind their decisions. The lack of transparency of such systems can have severe consequences or poor uses of limited valuable resources in medical diagnosis, financial decision-making, and in other high-stake domains. Therefore, the issue of ML explanation has experienced a surge in interest from the research community to application domains. While numerous explanation methods have been explored, there is a need for evaluations to quantify the quality of explanation methods to determine whether and to what extent the offered explainability achieves the defined objective, and compare available explanation methods and suggest the best explanation from the comparison for a specific task. This survey paper presents a comprehensive overview of methods proposed in the current literature for the evaluation of ML explanations. We identify properties of explainability from the review of definitions of explainability. The identified properties of explainability are used as objectives that evaluation metrics should achieve. The survey found that the quantitative metrics for both model-based and example-based explanations are primarily used to evaluate the parsimony/simplicity of interpretability, while the quantitative metrics for attribution-based explanations are primarily used to evaluate the soundness of fidelity of explainability. The survey also demonstrated that subjective measures, such as trust and confidence, have been embraced as the focal point for the human-centered evaluation of explainable systems. The paper concludes that the evaluation of ML explanations is a multidisciplinary research topic. It is also not possible to define an implementation of evaluation metrics, which can be applied to all explanation methods.