The safe operation of the smart grid system and the financial viability of the grid depends on increasing the detection rate of users who steal electricity in smart grids. Manual home inspection is the traditional method of detecting electricity theft, which is expensive and ineffective. Machine learning-based methods to identify electricity thieves are gradually being used with the usage of smart meters. In this paper the first step is Data pre-processing which includes removal of missing values, interpolation and data cleaning. There after supervised machine learning algorithms like Decision tree, K-nearest neighbor, and Random forest has been applied on a large data set for identification of non-technical losses. The experiment results on the real world electricity dataset show that K-NN performance is more efficient than other methods. Evaluation metrics like accuracy, recall, precision, and F1-score have been calculated. Overall, the paper highlights the potential of machine learning algorithms, particularly K-NN, in enhancing the detection rate (DR) of electricity theft in smart grids. It emphasizes the importance of data preprocessing and showcases the effectiveness of the proposed methodology through comprehensive evaluation and performance analysis.