A Data-Driven Air Quality Prediction Framework…

Air quality forecasting in semi-urban Indian regions remains significantly under-studied despite rising environmental risks and increasing public health concerns. Ahilyanagar, a rapidly developing district in Maharashtra, lacks transparent, downloadable, and continuous AQI data—resulting in decisions based on limited monthly classifications rather than granular pollutant measurements. In this study, I propose a hybrid analytical–predictive framework capable of utilizing categorical AQI records—representing the number of days per month falling under Low, Moderate, Poor, or Hazardous air-quality bands—from 2021 to 2025.

The model design includes three components: (1) extraction and structuring of historical categorical datasets, (2) visualization and seasonal pattern identification, and (3) development of a predictive engine using classical machine-learning techniques tailored for low-resolution data, including logistic regression, decision-tree classification, and baseline time-series models. Despite the absence of pollutant-level values (PM2.5, PM10, NO2, SO2, etc.), the proposed methodology demonstrates that useful trends and forecasts can still emerge from high-level classification data.

This work highlights the feasibility of deploying accessible forecasting tools for semi-urban municipalities, enabling early health advisories, better urban planning, and increased environmental awareness. The project also reinforces the role of student-led data science initiatives in bridging environmental information gaps in India. Future expansion—including API integrations, satellite-based feature engineering, and sensor-level monitoring—can further improve predictive reliability.

印度半城市化地区的空气质量预报研究仍显著不足，尽管当地环境风险不断攀升、公众健康顾虑与日俱增。马哈拉施特拉邦快速发展的艾哈迈德讷格尔（Ahilyanagar）地区目前缺乏透明、可下载且持续更新的空气质量指数（AQI, Air Quality Index）数据，导致决策仅依托有限的月度分类结果，而非精细化的污染物浓度监测数据。

本研究提出一种混合分析-预测框架，可利用2021年至2025年的分类式AQI记录——即每月分别属于低、中等、差、有害空气质量等级的天数——开展相关研究。

模型设计包含三个模块：（1）历史分类数据集的提取与结构化处理；（2）可视化与季节模式识别；（3）针对低分辨率数据定制的经典机器学习预测引擎开发，涵盖逻辑回归、决策树分类及基准时间序列模型。尽管缺乏PM2.5、PM10、NO2、SO2等污染物浓度数值，所提方法仍证明，从高层级分类数据中仍可挖掘出有价值的趋势与预测结果。

本研究验证了为半城市化市政机构部署易用型预报工具的可行性，该工具可助力发布早期健康预警、优化城市规划并提升公众环保意识。本项目同时凸显了学生主导的数据科学举措在填补印度环境信息缺口方面的重要作用。未来可通过集成应用程序编程接口（API, Application Programming Interface）、基于卫星的特征工程及传感器级监测等方式进一步提升预测可靠性。