Scholarlink

The high incidence and spatiotemporal variability of lightning discharges in tropical regions—such as Colombia’s Orinoquía—present major challenges for prediction and climate risk management. This research evaluates and compares traditional machine learning models (Random Forest, XGBoost, Support Vector Machine, Logistic Regression, and Naïve Bayes) with hybrid deep learning architectures (Multilayer Perceptron, Convolutional Neural Network, Feedforward Neural Network, Long Short-Term Memory, CNN-LSTM, and LSTM-MLP) to classify lightning occurrences within this region. Data preprocessing involved techniques such as spatiotemporal synchronization, undersampling, and normalization. The results revealed a significantly superior performance by the hybrid deep learning models, with the CNN + LSTM architecture achieving the best results: 79.0% accuracy, F1-score of 0.7994, and AUC-ROC of 0.7903.

错误：搜索内容不能为空，请输入英文关键词
错误：关键词超出字数限制，请精简
高级检索

Predictive Model Based on Thunderstorms in Colombia Using Machine Learning Techniques

  • 会议论文

  • 发表日期：

    2026年

查看全文
  • Briyid C. Cruz,
  • Paulo A. Gaona

摘要

The high incidence and spatiotemporal variability of lightning discharges in tropical regions—such as Colombia’s Orinoquía—present major challenges for prediction and climate risk management. This research evaluates and compares traditional machine learning models (Random Forest, XGBoost, Support Vector Machine, Logistic Regression, and Naïve Bayes) with hybrid deep learning architectures (Multilayer Perceptron, Convolutional Neural Network, Feedforward Neural Network, Long Short-Term Memory, CNN-LSTM, and LSTM-MLP) to classify lightning occurrences within this region. Data preprocessing involved techniques such as spatiotemporal synchronization, undersampling, and normalization. The results revealed a significantly superior performance by the hybrid deep learning models, with the CNN + LSTM architecture achieving the best results: 79.0% accuracy, F1-score of 0.7994, and AUC-ROC of 0.7903.