In the digital economy, detecting counterfeits in financial transactions is critical due to the rise in online transactions and sophisticated fraud techniques. The high volume and speed of transactions generate challenges, but machine learning is vital to detect counterfeits. This paper presents a bidirectional long-short-term memory (LSTM) architecture with an attention layer to enhance fraud detection. The dataset is balanced by the Synthetic Minority Oversampling TEchnique (SMOTE) and forward feature selection with a random forest classifier to select highly correlated features from the dataset. The model achieved at least 99.47% accuracy. It outperforms the convolutional neural network (CNN), the gated recurrent unit (GRU), and the LSTM with attention in terms of performance parameters, accuracy, precision, recall, F1 score, and receiver operating characteristic—area under the curve (ROC-AUC), demonstrating its effectiveness in detecting financial fraud.