Master's Thesis 2024 30 ECTS

Faculty of Science and Technology

Investigating the Viability of Mchine Learning for the Prediction of Icing Occurrences at Airports

Tonje Martine Lorgen Kirkholt

This repository contain the necessary code to get an overview of the work and processes explained and presented in the thesis. Files with code for extraction of data can not be run, as it is used internally at MET. Due to the final size of the dataset, it was not possible to upload the file to GIT. The dataset can be accessed and downloaded via this link: https://zenodo.org/records/11198572?token=eyJhbGciOiJIUzUxMiJ9.eyJpZCI6IjZhNDNmYThkLTNlYjEtNDI2MC1iYzZiLTEyNWNiMDM3MTgzZiIsImRhdGEiOnt9LCJyYW5kb20iOiI5MTQ1ODBiYmYwZjMyZDE0OWJkZWJjZTEwZDAwMDNlNCJ9.NAtiK_yL3r3Vklb0_gtAkIrxy6kT7Lz0jMggah8KvGiYaTK1FFj8pY4cb14feFN_0oNpF0_4-ARZ0PSxBWHzKg

The results in the thesis can be reproduced with this dataset and the models, remember to change the filepaths for importing the data in the notebook you want to run. Most of these notebooks are saved with a previous run, so results, statistics, and figures can be viewed as they are now.

An overview of the files and the different models, which can be located in 03_final_ml_training_test:

Persistence

• persistence_model: Majority class downsampled by 1%, dataset D1
• persistence_model_2: Majority class downsampled by 3.3%, dataset D2
• perstistence_model_without_AUTO: All instances of metartype AUTO are removed, dataset D3

Instantaneous

• instantaneous_1: Majority class downsampled by 1%, dataset D1
• instantaneous_3_3: Majority class downsampled by 3.3%, dataset D2
• instantaneous_without_AUTO: All instances of metartype AUTO are removed, dataset D3

Temporal

• lead_1_prediction: 1 timestep prediction = 1H ahead, dataset D1
• lead_1_prediction_2: 1 timestep prediction = 1H ahead, dataset D2
• lead_2_prediction: 2 timesteps prediction = 2H ahead, dataset D1
• lead_2_prediction_2: 2 timesteps prediction = 2H ahead, dataset D2
• lead_3_prediction: 2 timesteps prediction = 2H ahead, dataset D1
• lead_3_prediction_2: 3 timesteps prediction = 3H ahead, dataset D2
• lead_4_prediction: 3 timesteps prediction = 3H ahead, dataset D1
• lead_4_prediction_2: 4 timesteps prediction = 4H ahead, dataset D2
• lead_5_prediction: 3 timesteps prediction = 5H ahead, dataset D1
• lead_5_prediction_2: 5 timesteps prediction = 5H ahead, dataset D2

Files inside 00_data_preparation have been used for data extraction. Files inside 01_data_analyzation and 02_machine_learning show examples of data analyzation and preprocessing. Files inside data contains dataset containing only METAR information used early in the process.