Rainfall Prediction Using Machine Learning

• Published in: 2024 2nd International Conference on Disruptive Technologies (ICDT) pp. 582 - 588
• Main Authors: Bartwal, Kanchan, Pathak, Nilotpal, Alexander, John, Aeri, Manisha, Dhondiyal, Shiv Ashish, Awasthi, Shashank
• Format: Conference Proceeding
• Language: English
• Published: IEEE 15-03-2024
• Subjects: Atmospheric modeling; Buildings; Correlation; Data preprocessing; k-nn; Light Gradient Boosting; logistic regression; machine learning; Machine learning algorithms; Predictive models; Rain; rainfall; rainfall prediction; random forest; SVC; XGBoost
• DOI: 10.1109/ICDT61202.2024.10489249

Rainfall prediction is crucial across various sectors, and this research examines the effectiveness of machine learning (ML) algorithms in forecasting rainfall occurrences using meteorological data. The study rigorously explores a comprehensive methodology encompassing data preprocessing, model building with various ML algorithms, and thorough evaluation methods. The dataset consists of a broad array of meteorological variables, including temperature, humidity, wind speed, atmospheric pressure, and geographical features. The data preprocessing techniques included handling missing values using Theil-Sen regression, re-sampling for dataset balance, and direct mapping to encode categorical features. Exploratory data analysis involved using Seaborn and Matplotlib to visualize data imbalances, detect outliers using boxplots and explore feature correlations. The process emphasized feature engineering to refine model performance and dropped columns based on high correlation and irrelevance. The model building utilized a range of ML algorithms, including Random Forest, SVM, XGBoost, Logistic Regression, KNN, and LightGBM. Evaluation scores: precision, accuracy, F1 score, and recall were pivotal in assessing predictive performance. The study's findings showcased the effectiveness of ML algorithms, with particular attention to models such as Random Forest and XGBoost, which demonstrated high performance across both validation and testing sets. After meticulous evaluation, the final model selected for its superior performance in accurately predicting rainfall events was LightGBM. The study's findings reveal the effectiveness of ML algorithms, demonstrating remarkable accuracy and predictive power, particularly with models like Light GBM, Random Forest, and XGBoost, which scored high on both validation and testing sets. The comprehensive analysis provided valuable insights into the complex relationships within meteorological data and their predictability. This research showcases the potential of ML techniques in accurately predicting rainfall events, contributing to informed decision- making in various sectors reliant on weather predictions.