Lightning Strike Analysis Workflow

Data Loading and Preprocessing

• Reads lightning strike data from files in the llis_20250315 folder.
• Filters for cloud-to-ground strikes.
• Creates a datetime column for temporal analysis.

Feature Engineering

• Processes data in 1-minute time steps for each time window (1, 2, 3, 4, 5 minutes).
• Extracts features from a 10-minute historical window.
• Creates binary labels indicating whether a strike occurs in the next window minutes within the HK square.

Model Training

• Trains a separate Multi-Layer Perceptron (MLP) model for each time window.
• Uses features and labels specific to each window.
• Applies a train-test split tailored to each time window.

Evaluation and Prediction

• Evaluates each model on its respective test set.
• Saves each trained model.
• Prints predictions and accuracy metrics, including a detailed analysis of the first 10 samples for each model.

Output

• Displays the shape of features and labels for each time window.
• Reports test loss and accuracy metrics.
• Provides detailed prediction results for each time window.