Smart grid is an emerging platform adopted by many electric power utility companies to enhance proper service delivery as well as cost-effective operations. Automatic Fault Detection and Clearance (AFDC) is a part of intelligent technology initiatives established on Tanzania's grid aiming at detecting, managing, and handling fault with little or without human intervention. Being one of the components of AFDC, the Load Forecasting (LF) plays important role in feeding restoration and distributed energy resource agencies. However, the efficiency of the existing LF approaches is found to be compromised when it comes to data whose distribution is characterized by a random-walk behavior. Therefore, this research work proposes an efficient AFDC-based LF (LF-AFDC) which is capable of generating load demand profile based on fault data. Firstly, the design requirements of the LF-AFDC framework are established using focus group discussion and literature survey. Secondly, the design of the LF model is achieved through parameter calibration from the existing Multivariate Non-Linear Regression (MNLR) method. Thirdly, the design of the LF-AFDC framework is achieved based on the design requirements for the realtime and fault-driven forecasting. Findings indicate the capability of the proposed framework to forecast the next load profile from the given fault-date, fault-time, restoration duration, Gross-Domestic Product (GDP), number of customers, and population information. Furthermore, the simulation results indicate the capability of the extended MNLR (e-MNLR) method to outperform the ANN, SVM, KNN, MICE, MissForest and MNLR models.