Owing to the inherent complications in membrane distillation (MD) operations, it has become a challenge to acknowledge swiftly and appropriately to safeguard the quality of effluent, particularly when the processing cost is a prominent concern. Membrane wetting in MD operations is a major concern during long-term performance. In this study, machine learning (ML) methodologies were utilized to overcome the limitations of conventional mechanistic modeling. ML applications have never been explored to investigate how operational factors, such as water flux and salt flux, are affected during long-term MD performance. Furthermore, time-dependent factors were neglected, making it difficult to analyze the relationship between effluent quality and operational factors. Therefore, this study demonstrates a novel ML-based framework designed to enhance the performance of MD. The ML-based framework consists of an autoregressive integrated moving average (ARIMA) and utilizes a unique pathway to explain the impact of time series among operational factors. The accuracy of forecasting has been explored by utilizing 180 h (180 datasets), that was further used and divided into training (165 datasets) and test datasets (15 datasets). Eventually, the ARIMA model demonstrated a highly precise relationship order between the model and experimental data, which can be further used to forecast membrane performance in terms of wetting and fouling. The selected ARIMA model (3,2,1) appears to be an adequate model for water and salt flux data which has been effectively used to capture the course of permeate water and salt flux by producing the smallest forecast RMSE. The RMSE values were observed to be 0.22 and 0.05 for water and salt flux respectively, which can better predict long time series with high frequency. These frameworks can be applied for the early prediction of membrane wetting if ample high-resolution data are available. © 2022 The Institution of Chemical Engineers