Introduction to Automated Machine Learning (AutoML)

Automated Machine Learning (AutoML)

• AutoML is a supervised machine learning approach
• AutoML automatically explores and selects machine learning models
• AutoML compares the predictive performance of different algorithms
• Users still need to understand the elements involved in model development

Learning Objectives

• Define Automated Machine Learning
• Identify and compare various uses of automated modeling
• Investigate the Automated Machine Learning Process
• Summarize the value of ensemble models
• Construct and assess an Automated Machine Learning model

What is AutoML?

• Supervised learning has a defined target variable
• Unsupervised learning has no target variable
• Running each supervised technique individually and comparing accuracy results is time consuming
• AutoML is an efficient alternative

Questions That Might Arise

• How was the data collected and prepared?
• How did the model arrive at its conclusion?
• What is the blueprint of the model?
• Why did the model arrive at that conclusion?
• What variables impacted the predicted outcome?
• What patterns exist in the data?
• What are the reasons behind the recommended model?
• Are there data issues impacting the validity of the model?
• Is the model consistent in its predictions?
• Why is the model a good predictor?
• How accurate is the model?

AutoML in Marketing

• 40% of companies use machine learning to improve sales and marketing performance
• AutoML adoption rate is expected to increase significantly
• AutoML can be used for optimizing pricing strategy, forecasting demand, optimizing inventory, risk modeling, customer satisfaction prediction, customer acquisition, and reducing churn, efficient processing, predicting customer responses

Companies Using AutoML

• Facebook
• AirBnB
• Sumitomo Mitsui Banking Corporation (SMBC)
• Kroger
• The Philadelphia 76ers
• Blue Health Intelligence (BHI)
• United Airlines
• URBN
• Disney
• Pelephone
• Salesforce Einstein

Automated Machine Learning Process Steps

• Preparing the data
• Building models
• Creating ensemble models
• Recommending models

Data Preparation

• Handling missing data
• Handling outliers
• Variable selection
• Data transformation
• Data standardization
• Invalid data results in "garbage in, garbage out"
• Appropriate data preparation is crucial for accurate predictions

Model Building

• Many models are automatically built after the analyst specifies the dependent variable
• The purpose of a model is to extract insights from data
• AutoML uses pre-established modeling techniques accessible to all, from novices to experts

Creating Ensemble Models

• Combining different algorithms into a single "super model"
• This reduces issues like noise, bias, and inconsistent/skewed variance
• Ensemble models usually yield optimal predictive performance
• Understanding how different variables contribute to an outcome may be difficult

Simple Approaches to Ensemble Modeling

• Continuous target variables: average of predictions from multiple models
• Advanced technique: weighted average (higher quality data receives higher weight)
• Categorical target variables: majority rule (most common category)

Advanced Ensemble Methods-Bagging

• Bootstrap Aggregating (Bagging) uses two steps:
  • Step 1: Generate multiple random samples from the original dataset. The same data point can appear multiple times in the different samples.
  • Step 2: Train a model on each sample and combine the results by averaging the predictions for continuous outcomes, or by using the majority vote for categorical outcomes.

Advanced Ensemble Methods - Boosting

• Boosting aims to reduce model error by:
  • Observing errors in a model and oversampling misclassified records in the next model
  • Applying the model in steps
  • Repeatedly fitting a model to successively improving samples
  • Producing a final model with better performance than any individual model

Model Recommendation

• Multiple predictive models are evaluated
• The model with the most accurate predictions is chosen
• Accuracy is assessed based on how well the model identifies relationships and patterns to predict outcomes from new observations, not the original observations
• The most accurate prediction model(s) are used for improved decision-making

Case Study - Loan Data

• Lending Club aims to reduce loan default rates
• Supervised model identifies borrowers at high risk of default
• DataRobot (AutoML) identifies customers at greater risk of default
• Lending Club uses this information to target these customers with financial support programs

This quiz delves into the fundamentals of Automated Machine Learning (AutoML), covering its supervised learning approach and the processes involved in selecting and comparing different machine learning models. It aims to clarify the importance of understanding model development elements while efficiently assessing model performance. Test your knowledge of AutoML concepts and applications with this engaging quiz.