Due to the temperature dependence of numerous physical, chemical, and biological processes in rivers, stream water temperature (SWT) is a controlling indicator of river habitat. Accurate SWT estimation is essential for planning and managing water resources, aquatic ecosystems, and environmental impact assessments. Accurately predicting the temperature of stream water is difficult because many factors affect it, such as morphological, hydrological, and meteorological parameters. Because machine learning (ML) can model complex and nonlinearities between SWT and its predictors, it is being used more and more. This study shows that the adaptive network-based fuzzy inference system (ANFIS) model integrated with teaching–learning-based optimization (TLBO) algorithm can be utilized to predict precise SWT estimates. Two statistical indicators—root mean square errors (RMSE) and the index of agreement (IoA)—were considered to assess prediction accuracy of implemented models based on various input combinations involving the parameters for air temperature (Ta), precipitation (P), sediment flow (SF), and streamflow (Qf) of Shimoga station. During testing, the ANFIS-TLBO model (RMSE—6.4435, IoA—0.9546) performed better than the ANFIS model (RMSE—17.8804, IoA—0.8818). The outcomes demonstrated that TLBO was an effective addition to the standard ANFIS and could improve the model’s capability and dependability.