jkeun / automobile-regression Goto Github PK

• URL: https://archive.ics.uci.edu/ml/datasets/automobile
• Target: normalized-losses
• Missing values: denoted by quotation marks ('?'). Skip data samples with missing values in the target
• Features to ignore: 'symboling'

• Number of Instances: 205
• Number of Attributes: 26 total
  • 15 continuous
  • 1 integer
  • 10 nominal

1. Data Preparation

• remove 'symboling' feature
• change cat values to numeric ("num-of-doors", "num-of-cylinders")
• apply "?" to NaN
• remove samples with missing values in the target("normalized-losses"): 164 samples remained
• impute missing values in features ("num-of-doors", "bore", "stroke", "horespower", "peak-rpm", "price")
• divide "numeric" & "categorical" group

2. Data Split

• Outer CV = 5
• modeling set: 80%
  • Inner CV = 5
  • train: 80%
  • validation: 20%
• test set: 20%

3. Preprocessing

• Missing values: impute (knn)
• Real variables: scaling
• Categorical variables: dummies

4. Modeling

• Linear Models
  • Rdige
  • Lasso
  • Elastic Net
  • Polynomial
• Non-Linear Models
  • Kernel Ridge
  • SVR
  • Ensemble
  • Neural Network
  • Nearest Neighbors

5. Grid Search & Model Selection (train & validation)

• Pipeline
  • model
  • parameters range
  • scoring: neg_mean_squared_error
• Parameter Search
  • Scatter Plot (parameter X score)
  • best parameters
  • best score

5. Evaluation (test)

• Outer CV = 5
  • predictions
  • scores & mean score

• how to handle missing data (https://towardsdatascience.com/how-to-handle-missing-data-8646b18db0d4)
• scaling (https://datascienceschool.net/view-notebook/f43be7d6515b48c0beb909826993c856/)
• linear models (https://datascienceschool.net/view-notebook/83d5e4fff7d64cb2aecfd7e42e1ece5e/)