DEMON-NEUROHACK / Challenge-2-LA-Team-A

A data hackathon about predicting dementia status from data collected in India

Challenge-2-LA-Team-A (Demonic Flanders)

Anandita Nadkarni, Yash Mate, Ming Wang, Saurabh Koshatwar, Kanish Nimesh Shah, Jonah Fisher

1.Introduction

The goal of our project is to analysis the LASI-DAD data and learn the patterns to predict the whether a person has dementia.

2.Dataset

• LASI-DAD is the first and only nationally representative study on late-life cognition and dementia in India.
• It is drawn from what is currently a cross-sectional sample of 4,096 community-residing older adults 60+ years of age from the larger LASI study (N ~ 70,000), which is a prospective, multi-purpose population survey, representative of both the entire country and of each state within India.

3.Methods

Data Cleaning and Augmentation

• Handled Data Imbalance using SMOTE
• Data Imputation with K Nearest Neighbors
• Correlation analysis using Pearson
• Dimensionality reduction with:
  • *Removing the columns which have more than 50% missing values
  • *Removing the variables which have less than 5% correlation with the target
  • *Removing the variables which have more than 90% correlation with each other
  • *Using SHAP and Random Forest to select 27 most important features

Machine Learning

• Applied both classification and regression.
  • Highlight methods: Random Forest
    • NaN / missing values
    • Imbalanced Data
    • Robust to outliers
    • Reduces bias and variance
    • Avoids over fitting

4.Results

• Classification

• Regression