Machine Learning Assignment - Kaggle-style Competition

Overview

This assignment simulates a Kaggle-style competition with a dataset divided into train, dev, and test sets. The task is to create an accurate classifier, with the top classifiers receiving bonus points. The assignment consists of two main parts:

Part 1: Experimental Analysis (experiments.ipynb)

Objective

• Conduct preliminary data analysis to build and justify the selected model.

Tasks

• Exploration and preliminary data analysis.
• Justification for model selection.
• Demonstrate preprocessing steps and their impact on model accuracy.
• Describe hyperparameter search and its impact on model accuracy.

Part 2: Model Development (model.py)

Objective

• Build and submit a model using the provided template.

Tasks

• Adjust the template according to the chosen model from Part 1.
• Set base hyperparameters for the model.
• Allowed to add functions but must be confined to a single file submission.

Solution Overview

• The solution begins with importing necessary libraries and loading the dataset.
• Preliminary data analysis reveals an unbalanced dataset.
• LazyPredict is utilized to select the best algorithms for further adjustmemt.
• The dataset is balanced using over-sampling and under-sampling.
• Feature selection and data transformation techniques are applied.
• Various models are tested, compared, and adjusted using different parameters and features.
• Final models are validated against the 'dev' set to evaluate their performance.

File Descriptions

• experiments.ipynb: Jupyter notebook containing experimental analysis including data exploration, model selection, and hyperparameter search.
• model.py: Python script with the final model implementation, ready for submission.

How to Run

1. Load and run experiments.ipynb for experimental analysis and model selection.
2. Run model.py to train and test the final model using selected parameters from the experiment.

Dependencies

• Numpy
• Pandas
• scikit-learn
• XGBoost
• LightGBM
• imbalanced-learn
• seaborn
• LazyPredict

Conclusion

This assignment provides hands-on experience in Kaggle-style competitions, involving classifier selection, data preprocessing, model development, and hyperparameter search.