<![CDATA[The budgeting process for zakat distribution at BAZNAS is still conducted manually, creating risks of overbudgeting or underbudgeting due to fluctuations in assistance requests and variability in mustahik needs over time. This study aims to develop a data-driven forecasting model using a NARX Series-Parallel network optimized with the Levenberg-Marquardt (L-M) algorithm to support evidence-based zakat distribution and poverty alleviation planning at BAZNAS. The dataset consists of national zakat data for the 2022-2024 period, including Zakat Maal and Zakat Fitrah from 35 provinces interpolated into monthly timesteps (36 samples), and national ZIS-DSKL data for 2019-2024 interpolated into 72 monthly timesteps; all variables were normalized prior to modeling. The NARX Series-Parallel model was implemented for multi-step forecasting with performance evaluated using Mean Squared Error (MSE). The results show that the proposed model achieved low MSE values for Zakat Maal, Zakat Fitrah, and ZIS-DSKL forecasting, indicating high predictive accuracy under normalized data conditions, while model validation was conducted to mitigate overfitting and data leakage risks. This study contributes methodologically to the application of neural network-based time series forecasting in socio-economic data and highlights the limitation of interpolation-based datasets, while providing a practical quantitative basis for zakat policy planning and poverty alleviation strategies at BAZNAS in accordance with Law No. 23/2011.]]>
