Machine-learning models used in trading are typically fed engineered features rather than raw prices alone — and many of those features are long-standing technical indicators that predate AI by decades. Knowing what they measure helps demystify what a model is actually reacting to.
Moving averages (MA)
A moving average smooths price data by averaging it over a set number of periods (e.g., 50-day or 200-day), reducing short-term noise to show the underlying trend direction. A common signal derived from moving averages is a "crossover" — for example, a shorter-term average crossing above a longer-term one is often read as a bullish signal, and the reverse as bearish. Models frequently use multiple moving averages of different lengths as inputs to detect trend strength and direction.
Relative Strength Index (RSI)
RSI measures the speed and size of recent price changes on a scale of 0 to 100, commonly used to identify whether an instrument may be "overbought" (often above 70) or "oversold" (often below 30). It's a momentum indicator, not a prediction — an overbought reading means recent price gains have been strong and fast, not that a reversal is guaranteed to follow.
MACD (Moving Average Convergence Divergence)
MACD tracks the relationship between two moving averages of an instrument's price to help identify changes in momentum and trend direction, typically visualised with a signal line crossover similar in spirit to the moving-average crossover described above. It's widely used as a trend-confirmation input alongside other indicators rather than in isolation.
Bollinger Bands
Bollinger Bands plot a moving average alongside bands set a certain number of standard deviations above and below it, expanding and contracting with volatility. Price approaching or touching a band is sometimes used as an input suggesting the market may be relatively extended, though — like RSI — this reflects current volatility and positioning, not a forecast.
Volume-based indicators
Indicators that incorporate trading volume alongside price (such as On-Balance Volume) are used to gauge whether a price move is backed by strong participation or occurring on relatively thin volume, which some models weight as a signal of conviction behind a move.
What "AI" adds on top of these
A genuinely AI-driven system typically takes a combination of indicators like these — often dozens at once — as input features, and uses a trained model to learn which combinations and thresholds have historically preceded favourable outcomes, rather than relying on a single indicator's fixed rule (like "RSI below 30 = buy"). Whether that learned combination is capturing something durable or just fitting noise in historical data is exactly the question addressed in our guide on backtesting vs. forward testing.
None of these indicators, alone or combined, guarantee an outcome — they describe what has already happened in price and volume, which a model then uses to estimate probabilities about what might happen next. For the broader picture of how that estimate becomes a signal, see what AI trading signals actually are.