Abstract
Accurate demand forecasting is essential for retail operations as it directly impacts supply chain efficiency, inventory management, and financial performance. However, forecasting retail time series presents significant challenges due to their irregular patterns, hierarchical structures, and strong dependence on external factors such as promotions, pricing strategies, and socio-economic conditions. This study evaluates the effectiveness of Transformer-based architectures, specifically Vanilla Transformer, Informer, Autoformer, ETSformer, NSTransformer, and Reformer, for probabilistic time series forecasting in retail. A key focus is the integration of explanatory variables, such as calendar-related indicators, selling prices, and socio-economic factors, which play a crucial role in capturing demand fluctuations. This study assesses how incorporating these variables enhances forecast accuracy, addressing a research gap in the comprehensive evaluation of explanatory variables within multiple Transformer-based models. Empirical results, based on the M5 dataset, show that incorporating explanatory variables generally improves forecasting performance. Models leveraging these variables achieve up to 12.4% reduction in Normalized Root Mean Squared Error (NRMSE) and 2.9% improvement in Mean Absolute Scaled Error (MASE) compared to models that rely solely on past sales. Furthermore, probabilistic forecasting enhances decision making by quantifying uncertainty, providing more reliable demand predictions for risk management. These findings underscore the effectiveness of Transformer-based models in retail forecasting and emphasize the importance of integrating domain-specific explanatory variables to achieve more accurate, context-aware predictions in dynamic retail environments.

Abstract
This study investigates the integration of deep learning for single-modality and multimodal data within materials science. Traditional methods for materials discovery are often resource-intensive and slow, prompting the exploration of machine learning to streamline the prediction of material properties. While single-modality models have been effective, they often miss the complexities inherent in material data. The paper explores multimodal data integration—combining text, images, and tabular data—and demonstrates its potential to improve predictive accuracy. Utilizing the Alexandria dataset, the research introduces a custom methodology involving multimodal data creation, model tuning with AutoGluon framework, and evaluation through targeted fusion techniques. Results reveal that multimodal approaches enhance predictive accuracy and efficiency, particularly when text and image data are integrated. However, challenges remain in predicting complex features like band gaps. Future directions include incorporating new data types and refining specialized models to improve materials discovery and innovation.

Abstract
This study investigates the effectiveness of different hyperparameter tuning strategies for peer-to-peer risk management. Ensemble learning techniques have shown superior performance in this field compared to individual classifiers and traditional statistical methods. However, model performance is influenced not only by the choice of algorithm but also by hyperparameter tuning, which impacts both predictive accuracy and computational efficiency. This research compares the performance and efficiency of three widely used hyperparameter tuning methods, Grid Search, Random Search, and Optuna, across XGBoost, LightGBM, and Logistic Regression models. The analysis uses the Lending Club dataset, spanning from 2007 Q1 to 2020 Q3, with comprehensive data preprocessing to address missing values, class imbalance, and feature engineering. Model explainability is assessed through feature importance analysis to identify key drivers of default probability. The findings reveal comparable predictive performance among the tuning methods, evaluated using metrics such as G-mean, sensitivity, and specificity. However, Optuna significantly outperforms the others in computational efficiency; for instance, it is 10.7 times faster than Grid Search for XGBoost and 40.5 times faster for LightGBM. Additionally, variations in feature importance rankings across tuning methods influence model interpretability and the prioritization of risk factors. These insights underscore the importance of selecting appropriate hyperparameter tuning strategies to optimize both performance and explainability in peer-to-peer risk management models.

Abstract
Hyperparameter optimization (HPO) is critical for enhancing the predictive performance of machine learning models in credit risk assessment for peer-to-peer (P2P) lending. This study evaluates four HPO methods, Grid Search, Random Search, Hyperopt, and Optuna, across four models, Logistic Regression, Random Forest, XGBoost, and LightGBM, using three real-world datasets (Lending Club, Australia, Taiwan). We assess predictive accuracy (AUC, Sensitivity, Specificity, G-Mean), computational efficiency, robustness, and interpretability. LightGBM achieves the highest AUC (e.g., 70.77% on Lending Club, 93.25% on Australia, 77.85% on Taiwan), with XGBoost performing comparably. Bayesian methods (Hyperopt, Optuna) match or approach Grid Search’s accuracy while reducing runtime by up to 75.7-fold (e.g., 3.19 vs. 241.47 min for LightGBM on Lending Club). A sensitivity analysis confirms robust hyperparameter configurations, with AUC variations typically below 0.4% under ±10% perturbations. A feature importance analysis, using gain and SHAP metrics, identifies debt-to-income ratio and employment title as key default predictors, with stable rankings (Spearman correlation > 0.95, p<0.01) across tuning methods, enhancing model interpretability. Operational impact depends on data quality, scalable infrastructure, fairness audits for features like employment title, and stakeholder collaboration to ensure compliance with regulations like the EU AI Act and U.S. Equal Credit Opportunity Act. These findings advocate Bayesian HPO and ensemble models in P2P lending, offering scalable, transparent, and fair solutions for default prediction, with future research suggested to explore advanced resampling, cost-sensitive metrics, and feature interactions.

Abstract
This study investigates the effectiveness of Transformer-based models for retail demand forecasting. We evaluated vanilla Transformer, Informer, Autoformer, PatchTST, and temporal fusion Transformer (TFT) against traditional baselines like AutoARIMA and AutoETS. Model performance was assessed using mean absolute scaled error (MASE) and weighted quantile loss (WQL). The M5 competition dataset, comprising 30,490 time series from 10 stores, served as the evaluation benchmark. The results demonstrate that Transformer-based models significantly outperform traditional baselines, with Transformer, Informer, and TFT leading the performance metrics. These models achieved MASE improvements of 26% to 29% and WQL reductions of up to 34% compared to the seasonal Na & iuml;ve method, particularly excelling in short-term forecasts. While Autoformer and PatchTST also surpassed traditional methods, their performance was slightly lower, indicating the potential for further tuning. Additionally, this study highlights a trade-off between model complexity and computational efficiency, with Transformer models, though computationally intensive, offering superior forecasting accuracy compared to the significantly slower traditional models like AutoARIMA. These findings underscore the potential of Transformer-based approaches for enhancing retail demand forecasting, provided the computational demands are managed effectively.