https://github.com/fedesgh/asteorid_randomforest_classifier

Classifier model trained with unbalanced dataset ready for deployment
https://github.com/fedesgh/asteorid_randomforest_classifier

imblearn pandas pickle seaborn sklearn

Last synced: 4 months ago
JSON representation

Classifier model trained with unbalanced dataset ready for deployment

• Host: GitHub
• URL: https://github.com/fedesgh/asteorid_randomforest_classifier
• Owner: Fedesgh
• License: apache-2.0
• Created: 2024-08-31T17:24:46.000Z (almost 2 years ago)
• Default Branch: main
• Last Pushed: 2024-10-24T22:44:37.000Z (over 1 year ago)
• Last Synced: 2025-01-22T19:11:49.545Z (over 1 year ago)
• Topics: imblearn, pandas, pickle, seaborn, sklearn
• Language: Jupyter Notebook
• Homepage:
• Size: 13.8 MB
• Stars: 0
• Watchers: 1
• Forks: 0
• Open Issues: 0
• Metadata Files:
  • Readme: README.md
  • License: LICENSE

Awesome Lists containing this project

README

          
## The project

The goal of the project is to create a classifier with **high recall**, and pickle it in order to be ready for deployment.

The dataset was downloaded from Kaggle: https://www.kaggle.com/datasets/ivansher/nasa-nearest-earth-objects-1910-2024

We must build a predictive model able to detect hazardous asteroids, for its nature we are intereset in **recall** in other words we prefer false alarms instead to dont detect such dangerous asteroids.

## Models

We train using **GridSearchCV** severals models: **Kneboirghs** , **SVC** , **RandomForest**, **VotingClassifiers** 

Also we use **SMOTETomek** due to the imbalance of the data: **0.127 of the data are hazardous asteroid**, with a total data of 338166 rows.

![images/pairplot.png](images/pairplot.png)

## The best model

Our best model is **clf4** wich is **RandomForest** with 0.70 of recall and 0.82 of ROC_AUC

![images/modelsummary.jpg](images/modelsummary.jpg)