Search

Freshwater salinisation and alkalinisation strongly depend on human and natural factors. We used an explainable machine learning approach to investigate the impact of natural and human factors on the salinity and alkalinity in rivers in the United States. The model achieved coefficient of determination of 0.89 for salinity predictions and 0.92 for alkalinity predictions. Shapley additive explanations revealed salinisation is driven mainly by human factors like population density (18 %) and impervious surface percentage (13 %) and natural factors like run-off (14 %). The alkalinisation is mainly influenced by natural processes such as runoff (36 %) and soil-pH (20 %).

Conventional machine learning techniques in diagnosing cardiovascular disease have a limitation owing to the lack of interpretability of models. This study utilised an explainable machine learning approach to predict the likelihood of having CVD. Four machine learning models were employed for CVD diagnosis: Decision Tree (DT), K-Nearest Neighbor (KNN), Random Forest (RF), and Extreme Gradient Boost (XGB). Shapley Additive Explanations (SHAP) were used to provide reasoning for the models' predictions. Using these models and explanations, a user interface was developed to assist in diagnosing CVD. All four classification models demonstrated good accuracy in diagnosing CVD, with the KNN model showcasing the best performance (Accuracy: 71 %). SHAP provided the reasoning behind KNN predictions, and the predictive interface was developed by embedding these explanations to provide transparency behind the model's decisions.

Water quality assessment and prediction play crucial roles in ensuring the sustainability and safety of freshwater resources. This study aims to enhance water quality assessment and prediction by integrating advanced machine learning models with XAI techniques. Traditional methods, such as the water quality index, often require extensive data collection and laboratory analysis, making them resource-intensive. The weighted arithmetic water quality index is employed alongside machine learning models, specifically RF, LightGBM, and XGBoost, to predict water quality. The models' performance was evaluated using metrics such as MAE, RMSE, R2, and R. The results demonstrated high predictive accuracy, with XGBoost showing the best performance (R2 = 0.992, R = 0.996, MAE = 0.825, and RMSE = 1.381). Additionally, SHAP were used to interpret the model's predictions, revealing that COD and BOD are the most influential factors in determining water quality, while electrical conductivity, chloride, and nitrate had minimal impact. High dissolved oxygen levels were associated with lower water quality index, indicative of excellent water quality, while pH consistently influenced predictions. The findings suggest that the proposed approach offers a reliable and interpretable method for water quality prediction, which can significantly benefit water specialists and decision-makers.

This study introduces the first-ever self-explanatory interface for diagnosing diabetes patients using machine learning. We propose four classification models (Decision Tree (DT), K-nearest Neighbor (KNN), Support Vector Classification (SVC), and Extreme Gradient Boosting (XGB)) based on the publicly available diabetes dataset. To elucidate the inner workings of these models, we employed the machine learning interpretation method known as Shapley Additive Explanations (SHAP). All the models exhibited commendable accuracy in diagnosing patients with diabetes, with the XGB model showing a slight edge over the others. Utilising SHAP, we delved into the XGB model, providing in-depth insights into the reasoning behind its predictions at a granular level. Subsequently, we integrated the XGB model and SHAP’s local explanations into an interface to predict diabetes in patients. This interface serves a critical role as it diagnoses patients and offers transparent explanations for the decisions made, providing users with a heightened awareness of their current health conditions. Given the high-stakes nature of the medical field, this developed interface can be further enhanced by including more extensive clinical data, ultimately aiding medical professionals in their decision-making processes.

Chronic Kidney Disease (CKD) is increasingly recognised as a major health concern due to its rising prevalence. The average survival period without functioning kidneys is typically limited to approximately 18 days, creating a significant need for kidney transplants and dialysis. Early detection of CKD is crucial, and machine learning methods have proven effective in diagnosing the condition, despite their often opaque decision-making processes. This study utilised explainable machine learning to predict CKD, thereby overcoming the 'black box' nature of traditional machine learning predictions. Of the six machine learning algorithms evaluated, the extreme gradient boost (XGB) demonstrated the highest accuracy. For interpretability, the study employed Shapley Additive Explanations (SHAP) and Partial Dependency Plots (PDP), which elucidate the rationale behind the predictions and support the decision-making process. Moreover, for the first time, a graphical user interface with explanations was developed to diagnose the likelihood of CKD. Given the critical nature and high stakes of CKD, the use of explainable machine learning can aid healthcare professionals in making accurate diagnoses and identifying root causes.

Streamflow forecasting is crucial for effective water resource planning and early warning systems, especially in regions with complex hydrological behaviors and uncertainties. While machine learning (ML) has gained popularity for streamflow prediction, many studies have overlooked the predictability of future events considering anthropogenic, static physiographic, and dynamic climate variables. This study, for the first time, used a modified generative adversarial network (GAN) based model to predict streamflow. The adversarial training concept modifies and enhances the existing data to embed featureful information enough to capture extreme events rather than generating synthetic data instances. The model was trained using (sparse data) a combination of anthropogenic, static physiographic, and dynamic climate variables obtained from an ungauged basin to predict monthly streamflow. The GAN-based model was interpreted for the first time using local interpretable model-agnostic explanations (LIME), explaining the decision-making process of the GAN-based model. The GAN-based model achieved R2 from 0.933 to 0.942 in training and 0.93–0.94 in testing. Also, the extreme events in the testing period have been reasonably well captured. The LIME explanations generally adhere to the physical explanations provided by related work. This approach looks promising as it worked well with sparse data from an ungauged basin. The authors suggest this approach for future research work that focuses on machine learning-based streamflow predictions.

In this study, we used four different machine learning models - artificial neural network (ANN), support vector regression (SVR), k-nearest neighbor (KNN), and random forest (RF) - to predict the natural period of reinforced concrete frame structures with masonry infill walls. To interpret the models and their predictions, we employed Shapley additive explanations (SHAP), Local interpretable model-agnostic explanations (LIME), and partial dependency plots (PDP). All models showed good accuracy in predicting the fundamental period (T). The post-hoc explanations provided insights into (a) the importance of each feature, (b) their interaction, and (c) the underlying reasoning behind the predictions. For the first time, we have created a graphical interface that can predict the value of T along with its SHAP explanation. This interface can be useful in manually optimizing the design of reinforced concrete frame structures with masonry infill walls. However, the local explanations from SHAP and LIME exhibited significant discrepancies, and LIME underestimated the feature importance of dominant features compared to SHAP. These discrepancies observed in the explanations highlight the need for further research in the field of explainable artificial intelligence (XAI) in structural engineering.

The number of expressway road accidents in Sri Lanka has significantly increased (by 20%) due to the expansion of the transport network and high traffic volume. It is crucial to identify the causes of these crashes for effective road safety management. However, traditional statistical methods may be insufficient due to their inherent assumptions. This study utilized explainable machine learning to investigate the factors that affect the severity of traffic crashes on expressways. The study evaluated two groups of traffic crashes: fatal or severe crashes, and other crashes that included non-severe injuries or only property damage. Five factors that contribute to crashes were analyzed: road surface condition, road alignment, location, weather condition, and lighting effect. Four machine learning models (Random Forest (RF), Decision Tree (DT), extreme gradient boosting (XGB), K-Nearest Neighbor (KNN)) were developed and compared with Logistic Regression (LR) using 223 training and 56 testing data instances. The study revealed that the machine learning algorithms provided more accurate predictions than the LR model. To explain the machine learning models, Shapley Additive Explanations (SHAP) and Local Interpretable Model-agnostic Explanations (LIME) were used. These methods revealed that all five features decreased the possibility of occurrence of fatal accidents. SHAP and LIME explanations confirmed the known interactions between factors influencing crash severity in expressway operational conditions. These explanations increase the trust of end-users and domain experts on machine learning models. Furthermore, the study concluded that using explainable machine learning methods is more effective than traditional regression analysis in evaluating safety performance. Additionally, the results of the study can be utilized to improve road safety by providing accurate explanations for decision-making processes for black-box models.