Time series forecasting involves analyzing data points collected or recorded at specific time intervals to predict future values. Traditional machine learning approaches often use random data splitting, but time-based splitting can be more effective for time series data due to inherent temporal correlations and potential non-stationarity. Time series can be decomposed into components such as trend, seasonality, and residual noise, with models like additive and multiplicative used to represent these elements. Smoothing techniques, like exponential smoothing, enhance forecasting by reducing noise, while ARMA models leverage both autoregressive and moving average components to predict future values. The article emphasizes the importance of stationarity in time series data and explores methods like ARIMA for modeling, highlighting the significance of understanding autocorrelation and choosing appropriate model parameters.