Hive-Project: Taking advantage of AI for calculating predictive behavioural credit
At Hive Project we are taking benefits of Artificial Intelligence. One of the key features to our users is to take advantage of statistics and probabilistic modeling. Every investment opportunity is also a risk and in order to quantify that risk we are using few methods of calculation.
So of the aspects that are going to be used are related to bid data analysis of the pool of data available in the HVN network, such as:
- Do we know the borrower, what is the history with the borrower, analysis of social media connections and etc.
- What are the current market conditions, analysis of current market indicators
- Amount of ask and HVN network history
In order for calculating predictive behavioural cred we are going to use discriminative models: logistic regression. Linear regression is a linear classifier whose probability estimates calibration are integral part of the training algorithm.
Logistic regression is fowling in the category of probabilistic supervised learning algorithms.
Following picture represent odds vs model input:
Then we have that the odds ratio is:
odds ratio= (P(1))/(1-P(1))= w_0+ w_1 x_1+⋯+ w_n x_n
where the w_0,w_1 and so on presents model coefficients or weights. The variables and so forth are the inputs to the model.
Artificial Neural Network is the model that emulates behaviour of a neuron. When neural networks are used in predictive analysis there is specific network that is called multi-layer perception. In multi-layer perception neurons are organized in layers in fully connected, feed-forward network. Each neuron is simply a linear equation. So lets see how is that done. Starting from the logical regression equation:
odds ratio= w_0+ w_1 x_1+w_2 x_2+ …+ w_n x_n
and related diagram with f:X →Y
The neuron calculation function is presented as Single neuron:
and the multi layer perception calculation is presented as collection of neurons in multiple layers:
In order to train our model we are going to obtain relevant set of data for each market and train the model. This model is iterative learner, which means that model is learning through iterations rather then in single processing step. The leaning process is similar to how humans learn, through few iterations.
You can find more detailed description of Hive Project in our Whitepaper.