Efficient agricultural water management requires accurate determination of reference evapotranspiration (ETo). Artificial intelligence has recently managed to solve several water management challenges. Acquiring suitable data sets, choosing appropriate criteria, and identifying the optimal algorithm are significant challenges in machine learning. Machine learning may be an effective tool for estimating reference evapotranspiration using minimal meteorological inputs. This work aims to estimate the amount of ETo in the Khyber Pakhtunkhwa region of Pakistan, highlighted as an important agricultural hub, using machine learning. The machine learning techniques include support vector machines based on radial basis function in conjunction with a random forest, artificial neural network, and deep learning long short-term memory systems with 15 input combinations. The findings showed that all four strategies predicted ETo values with reasonable accuracy and reliability. The support vector regression model with scenario-14 revealed strong results than the others (R-2 = 0.9970). The ultimate purpose of this research is to develop the best scenarios with a reasonable accuracy rate and the least amount of input parameters to create the most suitable methodology. Hence, the second most successful models were artificial neural nets and deep learning with scenario-9 and 11, estimating ETo with a higher accuracy level (R-2= 0.9768 and 0.9688) during the testing stage. In conclusion, the research confirms the beneficial effects of support vector regression and enhanced performance of machine learning algorithms in estimating ETo in water shortage tropical and semiarid zones.