As urban flooding risks intensify due to climate change, developing accurate and real-time predictive systems has become critical for disaster preparedness and response. This study proposes an AI-based flood prediction model that leverages real-time rainfall and inflow water level time-series data. The system consists of two core modules: an anomaly detection and auto-correction module to ensure data reliability and quality, and a deep learning-based time-series forecasting module for predicting inflow levels based on rainfall trends. The anomaly detection module employs a reconstruction error method based on Long Short-Term Memory (LSTM) networks to identify abnormal data points that deviate from normal time-series patterns. Detected anomalies are corrected via linear interpolation, where the anomalous data points are removed and replaced with new values estimated from surrounding normal data points. The forecasting module, built upon the LSTM architecture, captures complex rainfall-runoff relationships and temporal dynamics to provide accurate water level predictions. Additionally, hybrid models including ConvLSTM and LSTM-Transformer were designed and evaluated for comparative performance. Experimental results demonstrated that the proposed system achieved over 95% accuracy in anomaly detection and correction. On real-world datasets, the LSTM model outperformed baseline methods, achieving a Mean Squared Error (MSE) of 0.000089, a Mean Absolute Error (MAE) of 0.006778, and a high R2 score of 0.972614. These results suggest the model's potential to enhance the reliability and efficiency of urban flood response decision-making and to contribute to the strengthening of disaster management capabilities.

With the rapid expansion of cloud computing, privilege abuse and escalation attacks in multi-cloud environments are emerging as a core security threat for organizations. Existing research shows limitations due to a lack of context from reliance on single data sources and the constraints of simple ensemble structures. To overcome these limitations, this study proposes a large-scale, multi-modal hybrid ensemble framework that integrates a total of 884,912 rows of data from the IBM Cloud Dataset, Microsoft Cloud Monitoring Dataset, and flaws.cloud CloudTrail data. The research methodology is as follows: (1) Large-Scale Data Integration: A total of 2,204,017 rows of raw data collected from three platforms were refined into an analysis dataset of 769,454 rows through time synchronization and 5-minute window aggregation. (2) Hybrid Ensemble Architecture: Isolation Forest, which is robust for outlier detection in high-dimensional data, and LSTM, specialized for learning sequential event patterns, are combined on a weight-based basis to simultaneously learn spatial and temporal anomaly patterns. (3) Multi-modal Feature Fusion: Infrastructure metrics, time-series service data, and privilege events were integrated to construct 15 standard feature vectors. In the experimental results, the proposed hybrid ensemble model recorded an ROC-AUC score of 0.721, showing superior overall performance compared to individual models. In particular, it demonstrated overwhelming performance in the Precision-Recall Curve analysis compared to other models, proving that it achieved a balance between reducing false positives and detection efficiency, which is crucial in real-world security environments. Furthermore, by explaining the model's prediction results through SHAP (SHapley Additive exPlanations) analysis, it secured transparency that allows security analysts to trust and act upon the findings.

While AI tutoring has strengths in personalization, it is limited by a lack of emotional and social interaction to keep learners engaged. To address this issue, this research proposes a new model that fuses AI tutoring with blockchain-based “token economy” and “gamification”. The token economy provides rewards for learning activities and community contributions (extrinsic motivation), while gamification provides the fun and sense of accomplishment of the learning process itself (intrinsic motivation). The synergy of these two elements aims to build a sustainable learning ecosystem by transforming learning into a “value-creating activity” rather than a mere obligation.

This study employs the X-Learner algorithm with deep learning models (MLP, LSTM, GRU, CNN) to causally estimate performance differences between cyclical (T=1) and defensive (T=0) asset classes in financial time-series data. The results show that the MLP-based policy achieved a Sharpe ratio of 0.440, while the LSTM-based policy improved to 0.990, outperforming T1 (0.880) and T0 (0.493). These findings indicate that incorporating temporal dependence through LSTM enhances causal policy learning. By reframing asset allocation as a causal decision-making problem rather than a predictive task, this study provides a foundation for interpretable and adaptive investment strategies in financial markets.

This study explores the enhancement of stock price prediction using StockGPT, a transformer-based generative model, by integrating macroeconomic indicators and feature engineering techniques. Using data from U.S. tech stocks and the S&P 500 index, the proposed model demonstrated approximately a 1.16% improvement in RMSE over the baseline. Feature importance analysis using Permutation Importance revealed that the lagged USD Index and the moving average of WTI oil prices significantly contributed to prediction accuracy. The findings underscore the practical impact of incorporating structured external variables into generative models and offer insights into interpretable financial forecasting.