?! IMPORTANT !? -> Problems downloading and compressing data, full documentation uploaded on drive and link-shared
We found it very challenging to do this challenge due to the complexity of data and being available to work with real world values, much more complex but interesting.
What it does
It predicts sale incomings for 3 weeks from now, the probability that a customer leaves the online buying process and we analyze some factors that may influence on this and some possibilities to fix it.
How we built it
Trying different Machine Learning / Deep Learning algorithms, like LSTM's and SVM, we built a time series predictor to obtain our results.
Challenges we ran into
Very difficult data to process, grouped by multiple-row users, high correlations that produce fake accuracies on training set and disasters when testing, and the difficult to process a time series related data.
Accomplishments that we're proud of
We've been able to do an implementation (at least a basic one, but very improvable with more working time) for a very complex problem without having deep knowledge on this topic.
What we learned
Deal with difficult and complex data
What's next for M.jar
Fix this project and try to improve the accuracy