I am familiar with many of the modeling techniques that you suggest, including logistic regression, decision trees, random forests and neural networks. Your proposed testing procedure is simply cross validation, using a portion of the data to develop the model, and the rest of the data to test it without re-fitting to the new data with the idea of seeing how extensible the model is. Model comparision would be done with the AIC or BIC. Finally, having chosen the best model, projections can be completed.
This work could be presented in two different ways at your choice:
1. A series of heavily commented R scripts
2. A knitr document with the R code embedded presented in either html (R Markdown) or pdf (Sweave) format.
Both formats are easily transported and can be verified on your end in either format.