Research and Data Science Repositories

I admit I am rather new in the research community and I have yet to publish any scientific papers on indexed journals. Most of my professional work involves fast, proactive research to be applied almost immediately for operational environments in dynamic marketplace conditions, which is as thrilling as it sounds if sloppy and as diametrically opposed to the scientific method as possible. Whatever work gets published is easily identifiable and attributed under my ORCid number. 


All relevant Data Science academic research can be found in my GitHub repository under MIT/GNU license for easy sharing. There are some interesting tidbits in my doctoral thesis on ensemble methods for FOREX prediction, especially for Spanish-speaking students. 


I use – and look forward to using more each day – RStudio’s Rpubs facility for sharing smaller documents and ideas that combine academics with R code. Whatever I publish here goes to the following repository.


Doctoral Thesis

The objective of my doctoral research thesis is the determination of a prediction model for the Colombian Peso exchange rate (legally known as the TRM or Market Representative Rate) using machine learning. The investigation focuses on the hypothesis that the time series representative of the TRM can be used as a prediction element for containing macroeconomic tendencies of the variable, yet the model becomes more robust when combined with the exogenous variables that intervene and enforce such behavior. For the TRM said independent variables are regressors representatives of the main export commodities for the country. Through machine learning data representing the different time series are used to generate an ARIMA forecasting model in function of the inherent fluctuation of the TRM, and a multivariable regression model using the TRM as the dependent variable and the different export commodities as independent regressors. Both trained models represent highly accurate forecasting tools whose results become the independent variables for a third ensemble model through stacking techniques, where the output of the first two learners become the input of a third with a higher level of accuracy.