The current judicial process is heavily influenced by the personal biases of the judges and officers related to the defendant's case. Because of this, there are many inconsistencies between the sentence time, parole offerings, and parole success for a given crime depending on who the defendant is. Factors like race, sex, and sentence type play a significant role in this system when the crime itself should be the primary influencer.
Data sets with information on parolees also hold information on other factors that may have influenced the parolee's action. A better understanding of the patterns within parolees' success will allow the prediction process in determining who should be offered parole more consistently. The goal of this project is to use specific factors to identify the probability of whether a defendant should be offered parole and how successful their time as a parolee will be.
If successful, this work could allow for predicting which defendants are suited for being offered parole. This type of prediction will lead clearer understanding of which defendants should be released for the parolee, excluding biases that are inevitably part of the current system.
We will use a Kaggle Dataset of Parole Violators to observe the following features for each song:
| male | race | age | state | time.served | max.sentence | multiple.offenses | crime | violator |
|---|---|---|---|---|---|---|---|---|
| 1 | 1 | 33.2 | 1 | 5.5 | 18 | 0 | 4 | 0 |
| 0 | 1 | 39.7 | 1 | 5.4 | 12 | 0 | 3 | 0 |
| 1 | 2 | 29.5 | 1 | 5.6 | 12 | 0 | 3 | 0 |
| 1 | 1 | 22.4 | 1 | 5.7 | 18 | 0 | 1 | 0 |
| 1 | 2 | 21.6 | 1 | 5.4 | 12 | 0 | 1 | 0 |
| 1 | 2 | 46.7 | 1 | 6 | 18 | 0 | 4 | 0 |
| 1 | 1 | 31 | 1 | 6 | 18 | 0 | 3 | 0 |
| 0 | 1 | 24.6 | 1 | 4.8 | 12 | 0 | 1 | 0 |
| 0 | 1 | 32.6 | 1 | 4.5 | 13 | 0 | 3 | 0 |
| 1 | 2 | 29.1 | 1 | 4.7 | 12 | 0 | 2 | 0 |
| 0 | 2 | 28.4 | 1 | 4.5 | 12 | 1 | 1 | 0 |
... | 1 | 2 | 35.4 | 1 | 5.3 | 8 | 0 | 3 | 0 |
Behavioral patterns are a good indicator of whether a parolee may violate their parole or if a dependent should be offered parole. Because the data set only has so much information, the prediction can only be as good as the data it is based on. Finding more data can help the model have a stronger prediction. Stigmatization for individual defendants or parolees can occur in the case of false negatives. To mitigate this, we can take out some factors that may have caused biases previously from skewing the data, removing the gender and race variables.
This is a regression problem: we will look at multiple variables and use those to predict the likelihood of a defendant's success on parole. Using many factors will allow for understanding which variables are most impactful and significantly influence the individuals.