Feature Selection and Classification on Madelon Dataset

Domain and Data

A synthetic dataset that was created based on UCI's Madelon dataset (https://archive.ics.uci.edu/ml/datasets/Madelon) This dataset contains 5000 features and 220,000 rows (1.1 billion data points) whereas only 50 features contain useful information

Problem Statement

This porject is to apply machine learning techniques to identify important features and then use these techniques to do prediction on the entire set.

Data Manipulation

Due to the size of the dataset, I had to built model on randomly selected subsets of the data.

• Subset 1: 0.1% of the data
• Subset 2: 0.5% of the data
• Subset 3: 1% of the data
• Subset 4: 5% of the data

Tasks

Featuer Engineering

• Use ANOVA for feature selection to select top 1000 features

Benchmark Modeling

• Build naive K-Neighbors and Logistic Regression benchmark models
• Use all 4 different size of data subsets

Feature Importance:

• Create feature selection (Select K Best/ PCA) pipelines
• Use feature selection / dimension reduction to reduce the dataset to a manageable size then use conventional methods

Implement Final Model:

• Present best model and apply to larger (or all) data subset

More Details can be found in /doc/MachineLearning_MadelonDataset.pdf