DataFest 2018 - Presentation by Team Quantum Banana
1. DataFest 2018 – Quantum Banana
Edward Chan. Mountain Chan. Ted MacCabe. Richard Watson. Yiqi Tang
Could Indeed data predict
unemployment?
Public Sector
How do I improve my job
posting?
Private Sector
2. Good/Bad Job
Posting Classifier
79%
- Clustered the data by unique jobs
- Weighted Local Clicks
- Calculated True Positive Rate
- Weighted unbalanced class
Data & Model
Top 25 percentile of job
postings in weighted clicks
are successful
We found our top 3 parameters
are Character Counts, Word
Counts and Estimated Salary
Top Influential Parameter
- Random Forest: 76% (TPP: 79%)
- Decision Tree: 70% (TPP: 70%)
- Neural Net: 80% (TPP: 20%)
- Consistent across DE, CA and US
Results
Private Sector
3. Public Sector
Trend of Monthly Unemployment Rate Forecast Model
Data & Model
• Clustered data by state and month
• Predicted unemployment change 3 months ahead
Future Improvements
• Adding in features from t-1
• Finding unemployment data with more granularity
Results
• Random Forest: 73%
• Decision Tree: 76%