Supervised Learning Techniques & Applications

Supervised Learning Defined

Supervised learning for machine learning occurs independent of the process. A supervisor is available while a machine learns how to complete tasks. After training completion, the expectation is that for new data sets the machine will be able to arrive at correct outcomes. As the number of practices on training data increase, outcome accuracy is expected to increase.

Supervised Learning Techniques

This approach is used when we have enough known data (labeled data) for outcomes that a model is attempting to predict. The learning designs an algorithm which maps inputs to outputs. The supervisory learning uses two techniques. The classification technique is appropriate when inputs can be segregated into categories. A real-world application is where the algorithm categorizes financial transactions as fraudulent or non-fraudulent. The regression technique takes inputs and predicts a single outcome for either continuous or real variables. A real world application would be using baseball Sabermetrics to predict how many games a MLB team will win.

Applications of Supervised Learning

Home Pricing – input data on square footage, number of rooms/bathrooms, yard size for predicting price
Face Recognition – input image to identify matches in surveillance footage
Weather Forecasting – current/historical data for predicting weather and precipitation
Customer satisfaction – sentiment analysis for classifying satisfied/dissatisfied customers
Recycling – robots sorting removing non-recyclable items from a conveyor waste stream
Television Viewing – recommending new viewing alternatives/degree of match based on past viewing
Customer Lifetime Value – determines net business profit of a specific customer over time
Marketing – scheduling, customizing and personalizing content for more effective marketing campaigns

Supervisory Learning Advantages & Disadvantages


  • Most appropriate for classification problems and predicting values from known data sets
  • More transparent approach compared to unsupervised learning
  • Complete control of the content of the input training data set
  • Classes and class boundaries are readily evident


  • Not appropriate for more complex machine learning tasks
  • Clustering cannot be completed based on input data features
  • Large input files can over train the model and distort model accuracy
  • Computation and classification processes are time-consuming efforts