Multi-task learning (MTL) is an approach where a single model learns multiple tasks simultaneously, potentially reducing overfitting and speeding up the learning process. However, selecting the right combination of tasks is crucial to avoid negative transfer, where learning additional tasks might hinder rather than enhance performance. Research has explored task relationships through shared data samples, features, relational knowledge, and parameters. However, sharing relational knowledge, features, or parameters often requires domain knowledge, potentially introducing artificial connections between unrelated tasks. This paper presents an MTL algorithm based on combinatorial U-curve optimization, focusing on data sample sharing. It uses a branch-and-bound-based approach to identify beneficial source tasks without requiring prior domain knowledge. The method organizes tasks into a tree, optimizing selection with backtracking and pruning for efficient exploration and minimal negative transfer. Results on synthetic and real-world datasets show that our algorithm is competitive with more sophisticated MTL frameworks such as Logistic L21, Dirty, and rMTFL, while maintaining a simpler design and fewer assumptions. In the Spam (Landmine) dataset, our algorithm yielded a mean loss that was 45% (62%) of the mean loss yielded by the second-best approach, Logistic Dirty (Logistic rMTFL). Although numerical improvements over these baselines are not always large, the proposed approach consistently matches or slightly outperforms them in most tasks, demonstrating robust performance even in scenarios with limited data or outliers.

Explainability is crucial for improving the transparency of black-box machine learning models. With the advancement of explanation methods such as LIME and SHAP, various XAI performance metrics have been developed to evaluate the quality of explanations. However, different explainers can provide contrasting explanations for the same prediction, introducing trade-offs across conflicting quality metrics. Although available aggregation approaches improve robustness, reducing explanations’ variability, very limited research employed a multi-criteria decision-making approach. To address this gap, this paper introduces a multi-criteria rank-based weighted aggregation method that balances multiple quality metrics simultaneously to produce an ensemble of explanation models. Furthermore, we propose rank-based versions of existing XAI metrics (complexity, faithfulness and stability) to better evaluate ranked feature importance explanations. Extensive experiments on publicly available datasets demonstrate the robustness of the proposed model across these metrics. Comparative analyses of various multi-criteria decision-making and rank aggregation algorithms showed that TOPSIS and WSUM are the best candidates for this use case.

Analyzing the impact of socioeconomic and urban variables on crime is a complex data analysis problem. Exploring synthetic, correlation-based scenarios using changes in a set of variables could alter a region's definition from unsafe to safe (known counterfactual explanation), which can aid decision-makers in interpreting crime in that region and define public policies to mitigate criminal activity. We propose CounterCrime, a visual analytics tool for crime analysis that uses counterfactual explanations to add insights for this problem. This tool employs various interactive visual metaphors to explore the counterfactual explorations generated in each region. To facilitate exploration, we organize our analysis at three levels: the whole city, the region group, and the regional level. This work proposes a new perspective in crime analysis by creating "what-if" scenarios and allowing decision-makers to anticipate changes that would make a region safer. The tool guides the user in selecting variables with the most significant effect in all city regions. Using a greedy strategy, the system recommends the best variables that may influence crime in unsafe regions as the user explores. Our tool allows for identifying the most appropriate counterfactual explorations at the regional level by grouping them by similarity and determining their feasibility by comparing them with existing examples in other regions. Using crime data from São Paulo, Brazil, we validated our results with case studies. These case studies reveal interesting findings; for example, scenarios that influence crime in a particular unsafe region (or set of regions) might not influence crime in other unsafe regions.

Extracting and analyzing crime patterns in big cities is a challenging spatiotemporal problem. The hardness of the problem is linked to two main factors, the sparse nature of the crime activity and its spread in large spatial areas. Sparseness hampers most time series (crime time series) comparison methods from working properly, while the handling of large urban areas tends to render the computational costs of such methods impractical. Visualizing different patterns hidden in crime time series data is another issue in this context, mainly due to the number of patterns that can show up in the time series analysis. In this article, we present a new methodology to deal with the issues above, enabling the analysis of spatiotemporal crime patterns in a street-level of detail. Our approach is made up of two main components designed to handle the spatial sparsity and spreading of crimes in large areas of the city. The first component relies on a stochastic mechanism from which one can visually analyze probable×intensive crime hotspots. Such analysis reveals important patterns that can not be observed in the typical intensity-based hotspot visualization. The second component builds upon a deep learning mechanism to embed crime time series in Cartesian space. From the embedding, one can identify spatial locations where the crime time series have similar behavior. The two components have been integrated into a web-based analytical tool called CriPAV (Crime Pattern Analysis and Visualization), which enables global as well as a street-level view of crime patterns. Developed in close collaboration with domain experts, CriPAV has been validated through a set of case studies with real crime data in São Paulo - Brazil. The provided experiments and case studies reveal the effectiveness of CriPAV in identifying patterns such as locations where crimes are not intense but highly probable to occur as well as locations that are far apart from each other but bear similar crime patterns.

On binary classification, the goal of minimizing the false positive and false negative rates creates a conflict, being impossible to optimize both simultaneously. This challenge is even more significant on imbalanced classification datasets since an incorrect choice of the relative relevance of each objective on the optimization can lead to ignoring, or poorly learning the minority class. The proposal of this work takes into account the existing conflict among the learning losses of the classes, and use a deterministic multi-objective optimization method, called MONISE, to create a set of solutions with diverse misclassification tradeoffs among the classes. Since accuracy is no longer a proper criterion for imbalanced datasets, we had to resort to multiple criteria to report the performance: each classifier, proposed or competitors, was selected and reported using the same metrics. We used F1, kappa and g-mean for general evaluation of performance and Fβs (F1/16, F1/4, F4and F16) to emulate a shifting decision maker preference from precision to recall; all comparisons were made using a Friedman test with Finner posthoc test. However, when we take into account multiple metrics without any prior knowledge, it may become impossible to pinpoint the best method, since the evaluation criteria may also be in conflict. Again, to solve this, we resorted to a Friedman test with a non-dominated ranking. With this multi-criteria analyses, we conclude that explicitly considering multiple objectives on the optimization can guide to promising results.

Regularized multitask learning is explicitly interpreted hereas a many-objective optimization problem, dealt with a deterministic solver that properly controls the sampling of the Pareto frontier. Each objective function corresponds to the learning loss of a task, so that we have as many objectives as tasks. The obtained Pareto-optimal models are then explored to implement distinct learning sharing strategies: (1) by considering a single parameter vector for all tasks, the simplest learning model that could have been conceived in multitask learning, the distinct trade-offs along the Pareto frontier can be interpreted as efficient and diverse sharing perspectives for the multiple tasks; (2) those distinct sharing perspectives are then aggregated in an ensemble or the best model in the validation set is selected. Notice that using a single parameter vector for all tasks in our many-objective perspective should not be directly associated with that naive, and generally of low performance, procedure of taking all tasks as being equally related. Distinct trade-offs automatically promote the proposition of efficient and structurally diverse relationships among the learning tasks, which support a competitive performance when compared with consolidated multitask learning methods in classification problems.