Feature Selection BTC-Dataset

Feature selection in a dataset with bitcoin data.

The goal of making feature selection on a dataset before placing it in front of any machine learning process is to:

• Reduce overfitting: If we have more columns in the data than the number of rows, we will be able to fit our training data perfectly, but that won’t generalize to the new samples. And thus we learn absolutely nothing.
• Occam’s Razor theorem: We want our models to be simple and explainable. We lose explainability when we have a lot of features.
• Garbage In Garbage out: Most of the times, we will have many non-informative features. For Example, Name or ID variables Poor-quality input will produce Poor-Quality output.

In this jupyter notebook i will analyze a bitcoin dataset created by myself but which I will not make available.

If you want you can build your own bitcoin dataset by looking at this repo: https://github.com/CristianCosci/bitcoin_OHLCV_dataset_generator and adding other data by referring to sites like https://www.blockchain.com/charts#currency and https://studio.glassnode.com/ through their API.

The dataset I will refer to will consist of the following features (Columns).

FEATURES : ['open', 'high', 'low', 'close', 'volume', 'new_addresses', 'active_addresses', 'total_unique_addresses', 'sending_addresses', 'receiving_addresses', 'total_fees', 'mean_transaction_fees', 'number_of_transaction', 'transaction_rate', 'market_cap', 'total_transaction_size', 'mean_transaction_size', 'hash_rate', 'mining_difficulty', 'reserve_risk', 'stock_to_flow', 'stock_to_fow_deflection', 'realized_profit', 'realized_loss', 'cvdd', 'address>001', 'addresses>1', 'exchanges_withdrawls', 'exchanges_deposits']

I will therefore select only the most useful features to predict the closing price of BITCOIN in USD.

The feature selection methodologies used are:

• Boruta
• Pearson Correlation
• Regression Feature Selection
• Mutual Information Feature Selection