- Formulate Business Problem. Identify what you are trying to solve. Also identify how are you going to measure the success of your solution
- Load the data set and identify the dependent and independent variables. Split the data into training and test set, if not done already
- Check for missing values, features with zero variance, outliers etc. Handle the missing values, analyze the outliers. Remove features that have zero variance.
- Do a visual inspection of data by drawing charts (heatmaps etc.)
- Identify the scale. Are all features at the same scale? If not, then look at scaling
- Identify the categorical features. Do OneHotEncoding for the categorical variables
- Too many features? Look at dimensionality reduction techniques like PCA/LDA
- Train the model on the training set
- Measure the outcome on training set
- Predict the values on the testing set
- Check for overfitting? If yes, look at L1/L2 regularization
- Repeat 8 -11 for other algorithms (for e.g. you can consider Logistic Regression, Decision Trees, Random Forest etc.)
- Evaluate Ensemble models, if necessary
- Identify the important features and try to understand why the model is predicting a specific output
