<![CDATA[This study analyzes the boiler machine maintenance system at PT. Karya Serasi Jaya Abadi using a predictive maintenance approach based on time series analysis with the ARIMA method to forecast critical component failure times and optimize maintenance schedules. Historical failure data over three years (2022-2024) from three critical boiler components—Superheater Tubes, Economizer Tubes, and Feed Water Pump—were analyzed using Mean Time Between Failures (MTBF) calculations, yielding an average of 339-340 operating hours between failures. Stationarity testing using Box-Cox transformation and Augmented Dickey-Fuller (ADF) test confirms data stationarity at the 5% significance level. Model identification using Autocorrelation Function (ACF) and Partial Autocorrelation Function (PACF) plots yields ARIMA(1,0,1) as the best model for all three components with the lowest Mean Square Error (MSE). Model validation through Ljung-Box Q-test and Anderson-Darling test confirms that residuals are white noise and normally distributed. Forecasting over 12 months shows stable failure intervals at 340-341 hours with 95% confidence intervals of 297-384 hours. ARIMA(1,0,1) model proves effective for predictive maintenance systems, enabling more accurate and planned maintenance scheduling, reducing unplanned downtime by 30-40%, enhancing operational reliability of the palm oil processing plant, and optimizing machine maintenance cost allocation. Keywords: predictive maintenance, ARIMA, boiler machine, time series, MTBF, machine reliability, maintenance scheduling]]>
