Extra Module 6 – Data Sciences Applied to finances.
Contrary to first 5 modules of this Alura Bootcamp Applied Data Sciences II (ABADSII), that were related to Covid-19 related matters, this extra module focused on data science applied to the financial market. More precisely the developpment of a machine learning model based on Vintage Analysis Theory. This model will then be run in an end-user web-based interface Credit Evaluator Simulator for our fictional bytebank.
In this module, we were guided by the Alura team throughout the complete life cycle of this Data Sciences project.
Overall, the project was developped in 3 phases using a large variety of tools:
-
Phase I - We covered from the basic, dataset import, wrangling, ETA, created functions, explored, and identified the best model for a credit evaluator simulator. All the work done on this phase used only one Google Colab notebook.
-
Phase II - We transfered to our hard drive the three key objets - fields' list, features and model, that closed the the phase I notebook. The design of these 3 objects was carefully thought to respond the tecnical needs of the the tools to be used in the phase 3 of our project
-
Phase III - We then became acquainted with the combined use of a triad of tools - Github desktop, Visual Study Code and Streamlit - to create and test a web-based interface to our credit evaluator simulator for our fictional bytebank.
AMF WARNING: I am rather uncomfortable with all the results obtained until the end of phase I. From my point of view all the progress of the process from one step to the other would require a stronger methodological rational and tied up. Particularly for the justification of the SMOT technique to ‘resolve’ a questionable unbalanced dataset (98% and 2%) for modelling. I only continued with the flow of the course because of the promised learning of techniques in phase II and III.
I am certain the overall proposal of this extra module 6 was lovable and conducted with the best of the intentions by the Alura team and I am thankful for that. I learned some interesting techniques that certainly will be used in other matters.
Nevertheless, the illusion of success brought by the final deliver, the web interactive interface, using a model based on a dataset, on my personal view, of questionable quality, is rather unsettling.
That said, I have no domain expertise on the subject at all, and would let to our bootcamp colleagues and/or anybody else with domainn expertise in finance and vintage analysis to opine on the matter.
Finally, I welcome any challenge and/or comments to opinions shared here above.
PS: I am keeping the title of this repository in portuguese as used in the curse, because I believe it is now the link between the 3 objects created on phase I and 3 differentes platforms used in the phase 3.