The client wanted to classify the bank transactions into pre defined categories using deep learning model to assign a two-level final classification for each valid input record. Based on labeled data set provided by client model has been developed.

• To devise a deep learning based model to classify bank transactions .
• Automated system has been developed using python environment as earlier they used to classify the each record manually into Level 1 and Level 2 categories.

• Data preprocessing has been done on description (input) column and was treated as a sequential data input .
• The length of text was limited so Character-Level Embeddings approach was found most suitable.
• CNN were used on character level embeddings. Other flag variables were trained using fully connected Neural network model and then combined using model merging.
• Deep Learning model was developed which can leverages both sequential data as well as cross sectional data.

The client implemented the model which was able to classify the top seven categories with accuracy of close to 100% and was also able to reduce the manual work in transaction classification process.