Human Guided Machine Learning Framework for Making Better Prediction

Abstract

Machine-made prediction is a key tool to gain faster performance rather than rely on hu- man prediction. Arti cial Intelligence showed better performance than human in many situations. That's why thousands of research have been taking place to develop more advanced Arti cial Intelligence which can not only perform faster but also predict better than human. But a human has some quality which can never be gained by a machine. Emotion, empathy, sensing, feeling are the characteristics where a machine cannot over- come human. These characteristics made human adaptive to take a decision over un- structured information, identify unusual circumstances and its consequences. So, we can say that a human and a machine both have their speci c quality in the certain scenario. In this paper, we will try to gure out if a machine can predict punishment like a human judge. We will implement a machine learning algorithm to create a system where human and machine can perform together to improve decision with less time. Additionally, the machine's performance will be checked by increasing the number of observations in a dif- ferent range. Through the study, we will try to evaluate the system whether the system can develop a new way to implement Arti cial Intelligence in the judicial system. In this research paper, we present a punishment prediction system where a human can give their decision also if necessary. We apply several machine learning algorithms such as Na ve Bayes, Logistic Regression, Support Vector Machine, Multiple Linear Regression, Arti cial Neural Network for Regression and Classi cation. By calculating both the test accuracy and the predictive power of the models, we observe which algorithm performs better and stable than the other models. We will also try to demonstrate what will be the impact of implementing such technology and what will be its technological, social, ethical and economic effect. In the end, we will try to state if a combined approach can produce a far more fruitful result than our regular judicial system. This combined model is a new way to represent an arti cially intelligent agent as a judge.