Big Data solutions for the forecast of PM2.5

Abstract

This thesis investigates the seasonal variation of PM2.5 concentrations in Debrecen, Hungary, and examines their relationship with meteorological factors. Using data from January and June 2024, the study analyzes correlations between PM2.5 and variables such as temperature, wind speed, and humidity. Principal Component Analysis (PCA) is applied to reduce dimensionality and extract key features from the data. These features are then used to train a Long Short-Term Memory (LSTM) neural network for forecasting PM2.5 levels. The model performs well in capturing winter trends but struggles with the low variability of summer data.