[tonyshaw]

Hi just want to get some initial feedback about the prototype.

We built a model based on all C--G historical loan records. The current algorithm is the gradient-descent boosting tree, we are planning to move to a deep neural network in the near future. Either way, the evaluation shows an 80+% accuracy in our test data set for predicting a charged-off loan. The model is quite selective, we think it will perform well. (Disclaimer: this is based on the historical result and may not reflect the truth).

The system works like this way:
1. On each day, it will check all listed loans, and select all loans that it thinks the borrower can fully-paid (loan only select from C-G).
2. Query your available cash, invest X loans that you have not invested by our system before. X is calculated by the min(number_selected_loan, avaliable_cash/25).

You are welcome to try this prototype and provide any feedback here.

Notice that:
The system is running on google app engine with free quota, and access availability is not guaranteed if there are too many users from this website.
The system will invest 25$ for each auto-selected loan. We are not able to control the total money to invest each order yet. Please control your available cash to below $250 just in case the model suddenly select tons of loans and drain all your cash

link:
http://investforests.appspot.com

You can check attachment for a sample of selected loans the model did for me.

[TravelingPennies]

I want to share some update about this auto selection.

I invest 250 usd per week and it has been going for 1+ month. Here is an initial collection I got.

I will keep you guys updated. Actually, I am not sure about the performance. But since this model is trained with GBDT with 85% acc on C-F notes, I am thinking this could have a better performance. I will keep you guys updated

[lascott]

from: tonyshaw on March 19, 2019, 01:49:54 AM