Please use this identifier to cite or link to this item:
|Title:||Financial forecasting using machine learning||Authors:||Chang, Hin Keat||Keywords:||Engineering::Computer science and engineering::Computing methodologies::Artificial intelligence
|Issue Date:||2022||Publisher:||Nanyang Technological University||Source:||Chang, H. K. (2022). Financial forecasting using machine learning. Final Year Project (FYP), Nanyang Technological University, Singapore. https://hdl.handle.net/10356/159318||Abstract:||Singapore’s economy is made up of 99% SME’s. They contribute to 70% of Singapore’s employment and 43% of the country’s Nominal Value. SME’s relatively simple structure allows them to respond quickly to the market as compared to larger corporations. On the flip side, their smaller size makes it difficult for SMEs to raise funds. Every business needs adequate capital for it to function. Financial forecasting can help calculate the financial needs of a business. Be it fixed capital or working capital, financial forecasting can make predictions on what a business needs to be successful. However, there are not any automated forecasting software in the market today. Prophet is an open-source forecasting algorithm created by Facebook in 2017. Facebook Prophet allows forecasting without the need for in-depth knowledge of time series coding and thus can be optimized by SMEs who might not have the necessary domain knowledge nor a deep understanding of time series data. We determined the optimum settings for prophet when dealing with both monthly and quarterly reported data. When using the quarterly and monthly seasonality in our prediction model reduces the errors into half and improves the r^2between the predicted and the actual by 10%. However, when analysing quarterly data having yearly and quarterly seasonality improves the accuracy. However, using forecasted operating cashflow as a regressor not only yields comparable result but also allows for the company to analyse its financial health via cashflow ratios. The automated parameter tuning is not effective for all companies and time frames. It yields a higher level of accuracy for models trained with 1 years’ worth of data, simulating start-up companies, with an improvement of 16%. While for models trained with 2 years of data or more it is more accurate without tuning the parameters.||URI:||https://hdl.handle.net/10356/159318||Fulltext Permission:||restricted||Fulltext Availability:||With Fulltext|
|Appears in Collections:||MAE Student Reports (FYP/IA/PA/PI)|
Updated on Dec 1, 2022
Updated on Dec 1, 2022
Items in DR-NTU are protected by copyright, with all rights reserved, unless otherwise indicated.