Weather Forecasting By Using Machine Learning

Abstract

Accurate weather forecasting is essential for numerous industries, including agriculture, transportation, energy, and disaster preparedness. Recent advancements in machine learning have sparked considerable interest in leveraging data-driven models to enhance weather predictions. This thesis delves into the application of various machine learning algorithms for weather forecasting, with a specific focus on developing an ensemble approach to optimize predictive accuracy. To achieve this objective, we employ the comprehensive London weather dataset, which spans a wide array of meteorological variables over an extended timeframe. Our study evaluates the performance of popular machine learning models, such as XGBoost, LSTM, GRU, and Holt Winter, using the Root Mean Squared Error (RMSE) as the primary performance metric. To explore the impact of different train-test split ratios on model performance, sensitivity analysis is conducted. This analysis sheds light on the optimal split ratio, with the 90-10 ratio emerging as the preferred choice. This knowledge serves as a pivotal component in our pursuit of developing an ensemble model that seamlessly integrates the strengths of XGBoost and GRU. Our ensemble model surpasses baseline XGBoost and individual models, demonstrating a significant improvement in predictive accuracy. The integration of XGBoost and GRU harnesses their complementary strengths, enabling the ensemble to capture complex patterns and temporal dependencies in the weather data effectively. Furthermore, we conduct feature importance analysis, which yields invaluable insights into the most influential variables driving weather patterns and predictions. Understanding these key drivers is crucial for gaining a deeper understanding of the underlying dynamics and improving the overall model performance.