pollybs / machine_learning_models_evaluation

Evaluation of the performance of machine learning models for predicting bankruptcies in Taiwan

(Évaluation de la performance de modèles de machine learning pour la prédiction de faillites à Taïwan)

Introduction

The goal of this project was to implement and evaluate supervised learning models using the Taiwan Bankruptcy Data. We created models for the prediction/classification and compared the metrics between the different models to determine the ones that performed better.

Models implemented:

• Logistic regression
• NuSVC
• BernoulliNB
• AdaBoostClassifier
• Linear Discriminant Analysis

Final Presentation

Steps

1. Descriptive analytics, data cleaning and formatting

2.Data preparation for machine learning and Feature selection:

• verification of correlation and
• reducing the number of variables,
• feature selection using VarianceThreshold , XGBoost, Kbest and Recursive Feature
• oversampling
• data scaling
• splitting the dataset (train/test)

3. Model research.

• Logistic regression
• NuSVC
• BernoulliNB
• AdaBoostClassifier
• Linear Discriminant Analysis

4. Implemented the models on our data

5. Hyperparameters tuning

6. Comparing the results using metrics:

• accuracy
• recall
• precision
• ROC_AUC score
• plot ROC_AUC curve