NHANES Age Prediction 🧠📊

In July 2025, I participated in the Summer Analytics Hackathon hosted by IIT Guwahati on AI Planet. The challenge was to predict whether a person belonged to the Adult (below 60) or Senior (60+) age group based on health and nutritional survey data from the NHANES dataset.

🏆 Hackathon Highlights

Hackathon Name: Summer Analytics 2025
Organized By: Consulting and Analytics Club, IIT Guwahati
Dataset: NHANES (National Health and Nutrition Examination Survey)
Goal: Classify individuals into Adult or Senior age groups.
Leaderboard Ranks:
- Private leaderboard: 176 / 220
- Public leaderboard: 160 / 236

🧑‍💻 My Approach

📝 Data Preprocessing

Cleaned missing values using mean imputation.
Converted categorical columns (e.g., Gender, Diabetes indicator) to numerical form.
Removed rows where the target variable was missing.

🛠 Feature Engineering

Created interaction features like:
- BMI x Glucose → to capture relationships between obesity & glucose levels.
- BMI x Insulin → to model insulin resistance tendencies.

🤖 Models Tried

Baseline: Logistic Regression.
Advanced Models:
- Random Forest (F1 ~30%)
- XGBoost (F1 ~35%)
- Stacking Ensemble (F1 ~37%)

🚀 The Stacking model combined XGBoost, Random Forest, and Logistic Regression to improve overall performance.

📈 Results

Model	F1 Score
Logistic Regression	18%
Random Forest	30%
XGBoost	35%
Stacking Ensemble	37%

📊 Top features (via XGBoost importance):

BMI
Glucose level
Insulin level
Interaction features like BMI x Glucose.

💡 Key Learnings

✅ Handling imbalanced classes using class_weight and scale_pos_weight.
✅ Importance of feature engineering in tabular data.
✅ How stacking can combine weak learners into a stronger predictor.

📜 Takeaways

This hackathon was a great hands-on experience in healthcare analytics. It taught me how to handle real-world datasets with missing values and imbalances and boosted my skills in ensemble learning.

🌟 “Achieved top 50% rank in a national-level hackathon by building an XGBoost-based age prediction model with advanced feature engineering and stacking ensemble techniques.”