Time Series + Classification: Background
Typical machine learning approaches in the past have been focused on two main areas: regression and classification. Regression tasks are usually performed on continuous data. Conversely, classification tasks usually consider some form of categorical data.
Time series data contain numerical values that are recorded at a consistent interval over time. Think of the values that your heart rate sensor records while you’re doing yoga or the daily stock price of your favorite cell phone company.
Regression Problems
A canonical example of a regression problem looks like this:
I have a 3 bedroom, two story home with 3 bathrooms and a swimming pool that was built in 2004. My zip code is 90210 and the home is 3,000 square feet. How much is it worth?
Here, we have numerical information, but it is not time series. The distinction here is that the thing we are attempting to predict is the value of the home. This is a continuous value and this is what makes the problem a regression task.
Classification Problems
It is important to point out that although classification problems focus on categorical data, they can still consider numerical information. For example, a canonical problem uses the Iris dataset and the objective is to classify three species of the flower based on the measurements of a few key phenomena. The distinction here is that these measurements can be viewed as a measure taken irrespective of time. Based on these features, some learning methods can be leveraged and we have classifications of each observation. From here, we can generate a number of performance metrics that help inform us of the model performance. These classification performance measures help to quantify how well the model predicts the correct class (or how poorly it tends to misclassify).
Time Series Classification Problems
Detail 4
Maecenas non leo laoreet, condimentum lorem nec, vulputate massa. Vivamus a ante congue, porta nunc nec, hendrerit turpis.