Supervised learning trains a model on labeled examples: features describe each case, while a known label or outcome supplies the prediction target. In financial research, the model can classify a stock as buy, sell, or hold, or estimate its future return. section 8.1
A practical supervised-learning workflow includes:
- Build an analytical table with an identifier, selected features, and one label or target outcome. section 8.1.2
- Choose quantifiable inputs such as prices, earnings, expenses, macroeconomic indicators, or other financial and nonfinancial data. section 8.1.2
Next-day price prediction
A compact experiment loads AAPL history with yfinance, keeps Open, High, Low, Close, and Volume, and shifts Close by one day to create Target. A RandomForestRegressor learns from the first 80% of observations, predicts the remaining observations, and evaluates them with root mean squared error. A chart compares the forecast with actual prices.
Supervised learning is useful but fragile in markets. Models can overfit or underfit, historical performance may not persist, and changing data sources, economic conditions, accounting practices, regulations, and industry patterns require caution and retraining. Trading applications must also account for slippage, transaction costs, and spreads. section 8.5 section 11.3.2
