Predict Customer Spending Using Machine Learning

machine learning

slug: recipe-machine-learning-predict-customer-spending-using-machine-learning

Recipe: Predict Customer Spending Using Machine Learning

category: business analytics / predictive modeling

Problem

You want to predict spending for new customers based on historical data so that you can:

• Estimate potential revenue per customer
• Identify high-value or low-value prospects
• Inform resource allocation or marketing strategies
• Evaluate model performance and refine predictive features

Solution

Train a machine learning model on historical customer data, then apply it to new customers to generate predicted outcomes:

• loadInline step — Define historical training data and new customer data inline
• trainModel step — Train a predictive model using selected features and target
• useModel step — Apply the trained model to new customers to predict outcomes
• print step — Display predictions for analysis or reporting

Step Sequence

loadInline step → trainModel step → loadInline step → useModel step → print step

Input Datasets

• Training dataset: Historical records including features and target variable

  • Features: numeric or categorical fields used for prediction (e.g., age, income)
  • Target: outcome variable to predict (e.g., spending)
• Prediction dataset: New records with feature values to score
• Both datasets must be structured as arrays of associative arrays (rows)

Output Dataset

• Original dataset enriched with predictions
• Key columns: all input features plus predicted_<target> (customizable via output_column)

Step-By-Step Explanation

Step	Purpose	Notes
loadInline step	Load historical training data	Inline data definition for small or test datasets
trainModel step	Train ML model	Select target, features, and modelType; store trained model in workflow extras
loadInline step	Load new customers	Define dataset for scoring
useModel step	Apply trained model	Predictions added as a new column (output_column)
print step	Display predictions	Can visualize predicted outcomes or export results

Variations & Extensions

• Train different model types (linear regression, logistic regression, k-NN) using modelType
• Normalize numeric features or handle missing values via parameters
• Split data into training and test sets using test_size for model evaluation
• Integrate predicted results with charts or dashboards for visualization
• Combine with filter step to score only a subset of customers

Concepts Demonstrated

• Training and applying machine learning models in a workflow
• Inline data definition for rapid prototyping
• Feature selection and predictive modeling parameters
• Appending predictions to new datasets for analysis

Related Recipes

• Benchmark segments against historical averages (Index)
• Classify customers by predicted outcomes for targeted campaigns
• Contribution + Pareto analysis to identify drivers of spending

Notes & Best Practices

• Ensure features in the prediction dataset match the training dataset exactly
• Normalize or preprocess features consistently between training and scoring
• Choose appropriate model type for target variable (numeric vs categorical)
• Use output_column to clearly label predicted results
• Optionally evaluate model performance using a test split or cross-validation

Example Workflow (DSL)

steps:
  - loadInline:
      data:
        - {age: 22, income: 38000, spend: 800}
        - {age: 25, income: 45000, spend: 1200}
        - {age: 29, income: 56000, spend: 1800}
      output: trainingData

  - trainModel:
      dataset: trainingData
      target: spend
      features: [age, income]
      modelType: linear
      params:
        normalize: true
        test_size: 0.2
        learning_rate: 0.01
        max_iterations: 1000
      output: spendModel

  - loadInline:
      data:
        - {age: 28, income: 52000}
        - {age: 42, income: 81000}
      output: newCustomers

  - useModel:
      dataset: newCustomers
      model: spendModel
      output_column: predicted_spend

  - print:
      title: "Predicted Customer Spending"

Example Output

age	income	predicted_spend
28	52000	1420
42	81000	2800