RociFi Scoring Model Update

Apr 12

We are thrilled to announce the release of the latest version of the RociFi scoring model, which incorporates several major improvements that will be beneficial to the end-users.

First and foremost, the model now utilizes transactional data from the Ethereum network, making it possible to score users who have never had a loan before with higher granularity. This means that if the user has good activity record on the Ethereum network, they can now obtain a favorable credit score, and their activity on the network will add bonuses to the final credit scoring.

Secondly, we have placed a greater emphasis on the transformation and selection of blockchain transaction features. By developing an activity coefficient that considers the number, frequency, and relationships between addresses, we have significantly enhanced the predictive qualities of our model. Additionally, our architecture is highly scalable, enabling the incorporation of new blockchain networks such as Polygon or BSC in future releases.

Thirdly, we have trained several models for each task and combined them into one ensemble. By combining the predictions of several models, we are able to reduce the variance in our predictions, leading to a more stable and accurate model. Moreover, we have employed various machine learning techniques, such as bagging and boosting, to train our models. Bagging involves training multiple models on different subsets of the data, while boosting involves training a sequence of models that each attempt to correct the errors of the previous model. Combining these techniques with the ensemble approach has resulted in a more powerful and accurate model that can better capture complex patterns in the data.

Lastly, we have changed the training pipeline of our machine learning models to improve their performance in predicting credit scores. We have worked extensively on finding the best hyperparameters for our models, which are the settings that control how the model is trained. By tuning these hyperparameters, we have made the model more adaptable to market changes and have improved its overall performance.

Additionally, we have made the model less conservative by adjusting the cutoff point for creditworthiness. Previously, the model was more stringent in assigning credit scores, but we have now adjusted it to be more forgiving. As a result, the distribution of credit scores has changed, and more users are likely to receive higher scores, reflecting their true creditworthiness. This change ensures that the model is fairer and more equitable, allowing more users to access credit and achieve their financial goals.

Below you will find a comparison of the distribution of credit ratings between the new and old version of the models.

Finally, we have also incorporated the recent increase in loan liquidations into our training pipeline. This means that our machine learning models now more accurately reflect the probability of loan defaults, allowing us to better manage credit risk and make more informed lending decisions. By adapting our models to reflect changing market conditions, we are ensuring that our customers can rely on the most up-to-date and accurate credit scoring system possible.

In conclusion, the release of the new RociFi scoring model marks a significant step forward in our ability to provide accurate and reliable credit scoring to our customers. By incorporating transactional data from the ethereum network and refining our feature selection and ensemble learning techniques, we have improved the stability, scalability, and accuracy of our model. We have also made the model less conservative, resulting in a fairer and more equitable distribution of credit scores. By incorporating recent market changes into our training pipeline, we are ensuring that our models remain up-to-date and relevant to the changing needs of our customers. Overall, these improvements reflect our commitment to providing the best possible credit scoring system to our customers and helping them achieve their financial goals.

Details of previous scoring updates you can see:

V2 Credit Scoring Update

TL;DR: In general, the majority of borrowers will see an improvement in their credit scores while some will slightly…

RociFi Credit Score Update — September 2022

First Announcement — Credit Score Model Update

RociFi — Capital-Efficient Lending Protocol on Polygon