This study analyzed the temporal variation and prediction of fine particulate matter (PM_2.5) concentrations in Aksaray, Turkey, a city in Central Anatolia. The relationship between PM_2.5 and meteorological parameters such as temperature, humidity, wind speed, and wind direction was investigated. An artificial neural network (ANN) model was developed to predict PM_2.5 levels based on meteorological data and air pollutant information. Seasonal and diurnal patterns of PM_2.5 concentrations were observed, with higher values recorded during the winter and lower values during the summer. Additionally, higher levels were observed in the morning and evening, while lower levels were recorded in the afternoon. The variations in meteorological parameters, especially temperature and wind speed, significantly influenced PM_2.5 levels. To predict hourly PM_2.5 concentrations, single and multiple data imputation techniques were employed in combination with resilient back-propagation (RPROP-ANN). The neural network was applied, consisting of one input layer comprising 11 parameters, one hidden layer with 20 neurons, and an output layer. The results indicate that the best forecasting performance for PM_2.5 was demonstrated by the combination of the missForest imputation technique with the RPROP neural network, as assessed by the coefficient of determination (R²), mean absolute error (MAE), and root mean square error (RMSE). The proposed model is characterized by a low RMSE of 5.94 and a high R² value of 0.88, demonstrating exceptional predictive performance in air quality.

中文翻译：

---

土耳其阿克萨赖每日细颗粒物 (PM2.5) 浓度预测：时间变化、气象依赖性和采用人工神经网络

本研究分析了土耳其阿克萨赖（安纳托利亚中部城市）细颗粒物 (PM _{2.5 ) 浓度的时间变化和预测。研究了PM 2.5}与温度、湿度、风速、风向等气象参数的关系。开发了人工神经网络 (ANN) 模型，根据气象数据和空气污染物信息预测 PM _2.5水平。观察到PM _2.5浓度的季节和昼夜模式，冬季记录的值较高，夏季记录的值较低。此外，早上和晚上的浓度较高，而下午的浓度较低。气象参数的变化，特别是温度和风速的变化，对 PM _2.5水平产生显着影响。为了预测每小时 PM _2.5浓度，将单数据插补技术和多数据插补技术与弹性反向传播 (RPROP-ANN) 相结合。采用的神经网络由一个包含 11 个参数的输入层、一个包含 20 个神经元的隐藏层和一个输出层组成。结果表明，通过将 missForest 插补技术与 RPROP 神经网络相结合，PM _2.5的最佳预测性能通过确定系数 ( R ² )、平均绝对误差 (MAE) 和均方根进行评估误差（RMSE）。所提出的模型的特点是 RMSE 较低，为 5.94，R ²值为 0.88，表现出卓越的空气质量预测性能。