Improved Commentary Prediction on Financial Data

Companies rely on financial reports which are generated through various transactions such as sales and expenses to understand the discrepancies between actual performance and financial forecast. Accordingly, generating commentaries on financial data might be considered as a routine operation for many companies. The previous studies indicate that machine learning algorithms can be used to automate the process of commentary generation. Specifically, such approaches use product forecasts and actuals in addition to inventory and point-of-sales data for the underlying prediction task. To incorporate these models in their daily activities, our partner proposes the development of a graphical user interface to allow end-users to interact with the model. By acquiring a deeper understanding of the data, we propose to improve the developed model in various ways. First, by engineering new features from the existing data, we aim to enhance the learning process and develop highly accurate models. Second, we plan to investigate time series classification approaches with deep neural networks considering that such methods enable pattern learning at different levels of detail. Finally, we consider applying natural language processing techniques to process commentaries and extract topics which provide a deeper understanding of the commentaries and serve as a validation tool.

Faculty Supervisor:

Mucahit Cevik

Student:

Sanaz Mohammadjafari

Partner:

Unilever Canada Inc

Discipline:

Engineering - mechanical

Sector:

Manufacturing

University:

Ryerson University