Kaggle’s Bike Sharing Demand

Solving Kaggle‘s ‘Bike Sharing Demand’ competition as a final project for

Naya college‘s Data Science Professional course instructed by Amit Rappel .

Project group members: Omer Vinik , Yechiel Hickry

Summary

1. Data Exploration

• Distinguish casual from registered customer by creating many custom graphs and plots:
  • Showing registered vs. casual rentals across day of the week
  • Showing registered vs. casual rentals across hour of the day
  • Showing average users count by weekday & hour of the day across user type
  • Showing registered vs. casual users count by hour of the day across weekdays
  • Showing registered vs. casual rentals by working day
  • Showing registered vs. casual rentals by season
  • Showing average temperatures by season
  • Showing registered vs. casual users count by hour of the day across seasons
  • Showing registered vs. casual rentals by weather
• Histograming the data and different targets
• Creating correlation analysis of the weather features

2. Feature Engineering / Data Manipulation

• Removing outliers from the weather features
• Extracting features from the Datetime column: Year, Month, Quater, Weekday, Hour
• Engineering new features from the Datetime column: isweekend, peak_hour, afternoon
• Engineering a new feature from the weather column: good_weather
• Scaling of 3 weather columns using StandardScaler
• Applying log(x+1) on the target columns - registered & casual counts

3. Execution

• Creating 2 different models for registered and casual customers
• Using 2 pipelines with custom transformers:
  • Add custom detail date and weather related columns transformer
  • Standard Scaler Transformer
  • Column Copier Transformer
  • Drop Column Transformer
  • Binned Data Digitizer Transformer
• Trying out many options of different features as input to the 2 models
• Applying RandomForest and XGBoost regressors using GridSearchCV

4. Validation

• Defining a RMSLE function and scorer
• Performing Cross Validation

5. Using test data for preparing competition submission