Predicting parole hearing result using machine learning

Machine learning is the science of getting computers to learn without being explicitly programmed. Tom Mitchell gives a formal definition of machine learning according to which, “A computer program is said to learn from experience ‘E’ with respect to some class of tasks ‘T’ and performance measure ‘P’, if its performance at tasks in ‘T’, as measured by ‘P’, improves with experience ‘E’ “. Thousands of parole eligible criminals are denied release every year. We need to consider what effect different factors have on the decision of parole commissioners so that we can know the probability of a prisoner being released from prison. We aim to train a machine learning model with training data of about 20,000 criminals in New York State that would then be able to predict whether a criminal would be granted parole or not. The model would take into consideration various attributes of each criminal like age, gender, race, charges, interview facility, etc. We assume that all these factors are considered by the parole commissioners when taking a decision. The raw data is taken and cleaned before it can be fed to a machine learning algorithm. We convert the string values of our data into numeric values by assigning a unique numeric value to each unique string value. Once we have a dataset with only numeric values, we normalize each attribute. T h e n we feed the normalized data to the Neural Network backpropagation algorithm and the graphical visualizations of the results are obtained using MATLAB. We are able to develop a model that can predict the result based on the attributes with an accuracy of 76.8%.