Criminal Recidivism is term which is use to mean someone who has committed some crime & does the same crime again after getting punishment & after coming out of prison that person is termed as Recidivist. ProPublica, a nonprofit news organization, had critically analyzed risk assessment software powered by AI known as COMPAS. COMPAS has being used to forecast which criminals are most likely to reoffend. However, when the algorithm was wrong in its predicting, the results was displayed differently for black and white offenders. Through COMPAS, black offenders were seen almost twice as likely as white offenders to be labeled a higher risk but not actually re-offend. To be precise there were gender and racial bias errors in the prediction.

What it does

Metaverse is tool use to reduce criminal recidivism by using machine learning classifiers and blockchain based evidence management system.

Machine Learning Classifiers

  • Using powerful machine classifiers, feature extraction is carried out on the compass dataset.
  • After feature extraction and preprocessing, K-cross validation is carried out on different powerful classifiers.
  • Based on the predictions, performance metrics would be computed to produce the best model.

Blockchain based evidence verification

  • Evidence manipulation is an important factor while dealing with false criminal recidivism system.
  • To overcome the above problem, we have proposed and implemented a blockchain based web portal to secure evidence files.
  • Immutability and traceability of evidence files would ensured through hashing in the blocks.

How we built it

The compass-two-years dataset has been utilized after feature engineering to train with different classifiers. Themis ML package was used to calculate mean difference and Confidence interval which helped us to extract Y variable(labels) and X & bias variables(Features).

  • The base model used is Logistic regression where all input variables (including protected attributes) - were used to train the set.
  • Remove Protected Attribute (RPA) was used without protected attributes in order to produce a naive fairness-aware approach.
  • Reject Option Classifier(ROC) and Added Counterfactually model (ACF) were used in the motive to produce better results.

Smart contracts are computerized ledgers intended to digitally facilitate, verify and enforce the contract specified through Ethereum virtual machine. The purpose of EVM is to compile and migrate our smart contract on Ethereum Blockchain. The use of smart contracts in our proposed system reduces convolution in the process by ensuring the evidence files would not be manipulated. To test our system we use tools like Truffle and Ganache-cli. Truffle is used to create, compile, migrate written smart contracts to running instances of the blockchain main network. Ganache is used to simulate client-based service and make the development process faster.

Challenges we ran into

  • Working on the blockchain
  • Date engineering
  • Selection of appropriate models

What we learned

For the bias to be treated, we are trying to make a trade-off between Fairness and utility According to our approach the mean difference should be low and accuracy should be high. From the above horizontal bar plot we can infer that for both LR and RPA classification models, both accuracy and mean difference are high. And with respect to ACF & ROC, the mean difference is low but it tends to lower the accuracy along with it. To conclude, ACF would be good classifier to the above problem as mean difference is significantly lower, but in terms of accuracy logistic regression would be a suitable classifier.

Blockchain technology, with the characteristics of transparency, security, immutability, accountability could contribute to reorganizing the evidence management system.

What's next for MetaVerse

  • Enhancement of model with fresh datasets.
  • Multi chain support for web portal.
Share this project: