Exploratory Analysis and Prediction of Weather Conditions: Leveraging Feature Engineering and Machine Learning Models for Accurate Forecasting

Abstract

This study explores the analysis and prediction of weather conditions using a comprehensive weather dataset containing variables such as precipitation, temperature, and wind speed. The research aims to enhance weather forecasting capabilities through data visualization techniques and machine learning models, addressing the critical need for accurate predictions in sectors like agriculture, transportation, and disaster preparedness. The research addresses gaps in the existing literature by focusing on localized weather patterns and exploring the potential of models that combine machine learning with traditional forecasting methods. However, challenges remain in model interpretability and adapting to climate change scenarios. Methodology comprises extensive feature engineering, for instance, building date-based features, temper-ature binning, and interaction terms. Multiple machine learning models, for example, Decision Trees and Support Vector Machines (SVM), classify the conditions of the weather based on engineered features. Metrics including accuracy, precision, recall, and F1-score are used for model evaluation. It is discovered that SVM have a tendency of outperforming other models. The paper also reveals feature engineering as vital in improving model performance, and it reveals that date-based features and interaction terms have played a significant role in improving prediction accuracy. It makes a research contribution in meteorology by demonstrating the effectiveness of machine learning for weather prediction and feature importance for weather classification. The findings have application in real-world practice for improving short-term weather prediction and decision support for weather-dependent industries.