Project Report

Analysis of ~80 million Indian district court data across states from the Development Data Lab.

The database consists of 3 types of folders

• Acts Sections metadata
• All cases metadata
• Judges metadata

The project was done fully on keggle and the dataset is also uploaded there , refer to the link

Insights on Data

1) Dependency of Various Columns of all cases data on duration a case lasted

• Used Mutual Info Regression to calculate MI score of various columns on duration of case
• Disp Name and Type name seemed to have high dependancy on duration .
• It is interesting to note that disp name is outcome of the case , but it still plays a role in predicting duration . Doesn't seem helpful ? Well , if you know a prisoner's trial duration in a court , especially when a prisoner applies for bail in higher courts , you know where to look !!

2) Tenure and Number of Judges in Different States

• Data divides states into 4 basic categories
  • T-1 states (pink) => Chandigarh , Telangana , Goa , Delhi
  • T-2 states (brown) => Assam , Bihar , Himachal , Jharkhand
  • T-3 states (orange,red) => Andhra , Karnataka , Kerela
  • T-4 states(blue) => Chattisgarh , Gujarat , Haryana , Mizoram
• Less populated states (T-1) tend to have a larger number of judges than fairly high populated states (T-3)
• Also , T-1 status being well-known to be "urban" or "educated" have larger male-female discrepancy than say , T-2 or T-4 states
• However , data suggests that judges in T-3 and T-4 states tend to have a longer tenure than judges in T-1 states . Does it indicate "youthfulness" or "Lack of Experience" in T-1 states ??

3) Combining All Cases Data and Judges Data to see cases per judge in various states

• I calculated cases/judge = (Total Number of Cases in a state)/(Total Number of Judges in that state)
• States with more cases / judge => Haryana , Karnataka ,Kerala , Chandigarh , Punjab
• States with less cases / judge => Jharkhand , Jammu and Kashmir , Manipur , Mizoram and Sikkim
• It is interesting to note that some states in here like Jharkhand and Jammu and Kashmir are densely populated and are not known to be "clean of crime" per se , but still they have very less cases per judge compared to other states . This makes us think that are all cases getting reported in these states ?? Or are these states really "crime free" contrary to popular opinion ??

Classifier

Predicting a Judge's Position Based on "State Code" , "Tenure" and "Start Date"

• Tenserflow environment with keras layers was used in this project
• Categorical Data was ordinally encoded and NaN values were filled with mean in data
• Data was split into training (70%) and Validation (30%)
• 3 Dense Layers were used with softmax activation in final layer
• Loss Function -> sparse_categorical_crossentropy , Optimizer -> adam and model was run for 10000 epochs
• Early stopping callback was also used to stop training when loss reaches its minimum
• Classifier outputs an array of 565 ( Number of Unique judge positions ) with each entry being the probability that the judge with the feeded input data (Start Date , Tenure and State Code ) holds the outputted position

How To Run

All the code blocks in the notebook are arranged in correct order of insights and Classifier. Copy the notebook and run the code blocks