Machine Learning: Logistic Regression

The Machine Learning: Logistic Regression indicator utilizes a classic classification algorithm to categorize market price movements into binary buy or sell states for trend analysis.

Usage

The Usage section describes how the script can be used to identify potential market entries and exits. The indicator retrains its model on every new bar, attempting to classify price action into two categories: Long or Short.

Signals are displayed as "Buy" and "Sell" labels on the chart. Users can also see "Exit" markers (crosses) based on a defined holding period. Because the algorithm relies on iterative weight adjustments, the interpretation of signals can be refined by observing the relationship between the price and the generated loss/prediction curves.

Example applications include:

• Trend Following: Using the classification labels to align with the current momentum.
• Mean Reversion: Observing when price deviates significantly from the scaled loss curve.
• Multi-Timeframe Analysis: Adjusting the "Resolution" setting to see predictions from higher timeframes on a lower timeframe chart.

Note: This indicator features repainting signals as it recalculates based on iterative training data.

Details

The script implements Logistic Regression, which, despite its name, is a classification algorithm rather than a standard regression tool. While linear regression fits a straight line to predict continuous values, Logistic Regression fits a Sigmoid (S-shaped) curve to separate data points into discrete classes.

The core execution involves a logistic_regression function that updates weights ($w$) via gradient descent over a user-defined number of iterations. The model seeks to minimize the log-loss between the prediction and the target data. In this implementation, the algorithm uses price data or temporal data to generate a probability of occurrence for a price move, which is then scaled using a Minimax algorithm to fit the chart's price scale.

Settings

Main Settings

• Price Type: Selects the source data for calculations (e.g., Close, HL2, OHLC4).
• Resolution: Determines the timeframe from which the data is pulled for the model.
• Lookback Window Size: Sets the window for the dot product and summation operations within the algorithm.
• Normalization Lookback: Defines the period used for the Minimax scaling of the output curves.
• Learning Rate: Controls the step size of the gradient descent during weight updates.
• Training Iterations: The number of times the algorithm loops to minimize loss and optimize weights.

Filters and Display

• Filter Signals by: Allows filtering signals based on Volatility (ATR), Volume (RSI of Volume), or both.
• Show Loss & Prediction Curves?: Toggles the visibility of the internal ML calculation lines.
• Optional Calculation?: Swaps the input variables for the logistic function (using time as a base).
• Use Price Data for Signal Generation?: Determines if signals are triggered by price crossing the curves or by curve crossovers.
• Holding Period: Sets the number of bars to wait before displaying an exit signal.
• Lot Size: Used for the internal backtesting statistics displayed in the info box.
• Show Info?: Displays a dashboard with cumulative return, trade count, and win rate.

FAQ

How do I use this indicator for different assets?

Some assets, such as BTCUSD or USDJPY, may require a lower "Normalization Lookback" (around 2–5) and enabling "Use Price Data for Signal Generation" for better alignment.

Why do the signals change after the bar closes?

This indicator uses an iterative machine learning process that recalculates on every bar, which leads to repainting. It is best used for historical analysis or with the understanding that signals may shift during live calculation.

How can I access the Machine Learning: Logistic Regression indicator?

You can get access on the LuxAlgo Library for charting platforms like TradingView, MetaTrader (MT4/MT5), and NinjaTrader for free.