Electrical theft is a global issue that harms both utility providers and electrical users. It destabilizes utility companies' economic development, creates electric dangers, and raises energy costs for customers. The development of smart grids is significant in power theft detection because they generate huge amounts of data, including consumer usage data, which may be used to detect electricity theft using machine learning and deep learning algorithms. This research presents a theft detection system that employs extensive information in the time and frequency domains in a deep neural network-based classification approach. We use data interpolation and synthetic data creation procedures to overcome dataset shortcomings such as missing data and class imbalance issues. Finally, we demonstrate the competitiveness of our strategy when compared to other methods assessed on the same dataset. During validation, we achieved 90% area under the curve (ROC), which is 1% greater than the best DNN in current works, and 94.48% accuracy, which is the second-highest on the benchmark.