Presented my master’s thesis work “Comparative Study of Deep Learning & Statistical Models in Time Series Forecasting” at the Bangladesh Statistical Association’s national conference, held under the theme “Advancing Official Statistics through Data Science: Bridging Tradition and Innovation”.
The paper aimed to answer whether for univariate multi-step forecasting of meteorological variables in Bangladesh, deep learning methods actually beat classical statistical models often enough to justify their cost?
The comparison covers rainfall, temperature and ultraviolet radiation at three climatically different sites, Chattogram, Dhaka and Sylhet, where seasonal monsoons and erratic weather make the forecasting problem genuinely hard rather than a benchmark exercise.