A training dataset is the labeled portion of a sample used to fit a predictive model. The algorithm processes features and labels to identify useful correlations. section 8.1.2
A practical supervised-learning workflow is:
- Collect and aggregate historical data over a time interval appropriate to the industry and prediction goal. section 8.1.2
- Split the labeled sample into training data and test data, keeping the test portion aside. section 8.1.2
- Fit a model on the training data, then evaluate its predictions against labels in the untouched test data. section 8.1.2
Stock-price prediction
A Python workflow loads Apple price history, retains Open, High, Low, Close, and Volume as features, creates the next Close as Target, and trains a random forest on the first 80% of observations. It evaluates the remaining observations with RMSE and compares actual and predicted prices in a plot.
