This paper presents a data-driven method for estimating the maximum Lyapunov exponent (LLE) from one-dimensional chaotic time series data using machine learning. The method generates multi-horizon forecasts using trained predictors and infers the LLE from the exponential growth of the geometric mean prediction error (GMAE). This approach is validated on four representative one-dimensional maps: logistic, sine, cubic, and Chebyshev maps, achieving an accuracy of R2pos > 0.99 even for short time series with M = 450. The best fit was achieved using KNN as a baseline. This estimator is designed for positive exponents and returns values indistinguishable from zero in periodic/stationary environments. Noise robustness is evaluated, with accuracy saturating at SNRm > 30 dB and degrading below 27 dB. The method is simple, computationally efficient, model-independent, and requires only stationarity and the presence of a dominant positive exponent. This method provides a practical way to estimate LLE in experimental settings where only scalar time series measurements are available, and its extension to high-dimensional and irregularly sampled data is left for future research.
