9. Regression

Regression is the task of predicting / estimating a continuous value. Training is performed using supervised machine learning, using a labeled dataset. It can be applied to standard a set of input values, or to time-series data. The output can be either a single data point for a time-series, or one per time-step.

9.1. Applications

Detecting events using Machine Learning has a wide range of applications, and it is commonly performed on sensor data using embedded systems.

Application examples

Area	Task	Sensor
Electronics	Battery power estimation	Voltage/current
Robotics	Distance estimation	Ultrasound
Health	Breathing rate estimation	Sound
Sensors	Calibration of air quality sensors	PM2.5 sensor
Industrial	Gas concentration estimation	Metal-Oxide semiconductor (MOS)
Industrial	Remaining Useful Life (RUL) estimation
Industrial	Prediction of failures using Time-to-Event	Accelerometer etc.

9.2. Classification models

Supported regression models

Algorithm	Implementation
RandomForest	RandomForestRegressor
ExtraTrees	ExtraTreesRegressor
DecisionTree	DecisionTreeRegressor
Multi-Layer-Perceptron	MLPRegressor, keras.Sequential

A basic example of some of the regressions models can be found in Regression models.

9.3. Related

Some specialized task formulations that use regression are: Time-series forecasting, for predicting future values, and Time-to-Event estimation, which is similar to to Event Detection.