The core objective of this research is to construct a machine learning-based air quality prediction model. This model aims to forecast the Air Quality Index (AQI) for the next 72 hours and classify its corresponding levels (e.g., Good, Moderate, Polluted), providing a robust scientific basis for environmental protection departments and related decision-making. For feature selection, we analyzed multiple key factors affecting air quality. While meteorological data, spatiotemporal features, and external pollution sources are important, this study focuses on the historical concentrations of six critical pollutants (PM2.5, PM10, SO2, NO2, CO, and O3) as model inputs to establish a baseline model, acknowledging the need for incorporating broader influencing factors in future work. In the model construction phase, we performed extensive preprocessing on the collected historical air quality data, including standardization and normalization, to extract effective information. We then employed and compared several advanced machine learning algorithms, selecting the optimal combination to build the final prediction model. The experiments were conducted using the Python language. By continuously optimizing model parameters and feature combinations, we achieved predictions for both the numerical AQI values and their corresponding quality levels for the subsequent 72 hours. Experimental results demonstrate that the constructed model possesses high prediction accuracy and stability for the predominant “Excellent” and “Good” categories. However, the lack of severe pollution events in the dataset limits the evaluation of its predictive capability for pollution peak events.
Shiting Wu, Xiaohua Qian, Xiaodong Zhou et al.· 2026 3rd World Conference on...· 0 citations
Large language model (LLM) applications deployed in regulated and high-risk environments face two coupled runtime risks: input-side prompt injection and output-side content compliance violations. Existing defenses are often either model-centric (alignment or fine-tuning) or application-local (keywords and regex), which limits auditability, policy agility, and cross-model portability. This paper proposes a gateway-centric framework that externalizes safety enforcement into a security side-chain. The framework integrates a unified LLM gateway with an external semantic guardrail service to perform bidirectional inspection before and after model invocation. We formalize the architecture with input-risk and output-compliance functions, define a comprehensive threat model and trust boundary, and present a reproducible evaluation protocol. To validate our approach, a prototype implementation based on LiteLLM and Prisma AIRS was evaluated on a rigorously balanced 2,000-sample adversarial dataset merged from diverse sources, including Deepset, wildjailbreak, and jailbreak-classification. Results demonstrate that the proposed architecture achieves an exceptional balance in detection capability, yielding an F1-score of 0.8322, a recall of 0.8531, and effectively maintaining a low False Positive Rate (FPR), thereby outperforming multiple state-of-the-art baselines including Llama Prompt Guard and Deepset DeBERTa. Furthermore, performance testing confirms practical deployability, showing an average added latency of only 320 ms under a 50-concurrent-user load. The paper concludes by discussing the side-chain advantages over model alignment mechanisms, guardrail bypass risks, and adaptive policy-evolution strategies for enterprise AI governance.