IRJET- Predicting Outcome of Judicial Cases and Analysis using Machine Learning

International Research Journal of Engineering and Technology (IRJET) e-ISSN: 2395-0056
Volume: 06 Issue: 03 Mar|2019 www.irjet.net p-ISSN: 2395-0072
© 2019, IRJET | Impact Factor value: 7.211 | ISO 9001:2008 Certified Journal | Page 326
Predicting Outcome of Judicial Cases and Analysis using Machine
Learning
Prof. Priyanka Bhilare1, Neha Parab2, Namrata Soni3, Bhakti Thakur4
1Assitant Professor, Dept. of Computer Engineering, Rajiv Gandhi Institute of Technology, Maharashtra, India
2Student, Dept. of Computer Engineering, Rajiv Gandhi Institute of Technology, Maharashtra, India
3 Student, Dept. of Computer Engineering, Rajiv Gandhi Institute of Technology, Maharashtra, India
4Student, Dept. of Computer Engineering, Rajiv Gandhi Institute of Technology, Maharashtra, India
---------------------------------------------------------------------***---------------------------------------------------------------------
Abstract – Today, the world is on the verge of cultivating
the most advanced and modernized civilization of mankind.
The progress in terms of architecture and technology is
enormous. The society has greatly evolved andthehumankind
is availing the most progressed, enlightened, cultured, refined
and sophisticated way of life. This remarkable attainment is
largely attributed to the field of technology and especially to
the field of law. Law and society form the most strong and
crucial establishment. The Judicial System has outreached
every aspect of the society and this inception is largely
expanding. But, In the recent times, thechallengesfacedbythe
civilians with the judicial system is becoming more
outrageous. The non-transparency and lack of knowledge of
the judicial proceedings has led to great disturbances and
uncertainty amongst the communities confined to a certain
legislature. This insufficient understanding of the legalsystem
has caused extensive deprivation intermsofrenumeration and
time. With the advancement in technology there has been
multiple attempts to assist the civilians in the judicialmatters.
This Project is in the similar endeavor to assists the civilians
with the legal matters and bridge the increasing gap between
the judiciary and the society. The emergence of Artificial
Intelligence and relative technologies made prediction of the
decision of a judicial system possible. This Project aims to
predict the outcome of a legal case with the maximum
accuracy.
Key Words: Artificial Intelligence, Machine learning,
Judicial Case, Outcome, Prediction, Civilians,
performance comparison.
1.INTRODUCTION
Law and society are the most inseparable attainments of
mankind. With the advent of human civilization began the
formulation of the judicial system. The legal system and
jurisprudence have the greatest significance in the
development of a society. The enforcement of law is pivotal
in modulating the growing communities. Due to the
prolonged expansion and modernisation in the society, the
theory and philosophy of law has also been subjected to
many curtailments and advances. Many communities and
societies do not include jurisprudence as a part of their
primary education. Because of this a large number of
civilians remain unaware of the legal proceedings.
A lawsuit can be very challenging especiallyforthosewho
are unfamiliar with the legal process which includes the
following major acts: meeting with the potential lawyers,
hiring a right lawyer, deciding in which court to file the
lawsuit and considering the applicable remuneration and
time period to sue. An uninformed civilian may fall prey to
any false claims proposed to him and may endure intense
expense in terms of time and funds. This system is
propounded with a motive to provide right guidance by
providing the appropriate prediction of the outcome of a
lawsuit or litigation in accordance to the judiciary system.
The system formulates this objective of legal foresee by
acquiring and working on the principles of Artificial
Intelligence. The beginning of Artificial Intelligence was
endowed with many myths, stories and was once quoted by
Pamela McCordick as he writes, “I beganwith‘a ancientwish
to forge the gods’ “. With the extensive research and study
being conducted, the world has onlooked the astonishing
applications of artificial intelligence.
Today, the applications of machine learning have
outstretched in all aspects of human life including
businesses, finance and medicine. This system is also the
follow up of data analysis and machine learning. There are
multiple research and implementations carried out in this
domain. Artificial Intelligence, especially in Natural
Language processing (NLP) and machine learning is gaining
quiet a attention of various problems. These technologies
provide us with tools to analyse the text and summons up
the semantics of the text involved. Overall, we believe that
building a text-based predictive system of judicial decision
can offer the civilians as well as lawyers the insight of their
law cases and help them analyse the outcomes of the case.
2. Predictive Models
A. Logistic Regression
Logistic regression algorithm can be used for both
classification as well as regression. Logistic regression is
similar to the linear regression with it logistic score. We use

Volume: 06 Issue: 02 | Feb 2019 www.irjet.net p-ISSN: 2395-0072
independent feature of the data to predict or classify the
dependent target value. This algorithm is similar to
perceptron neural algorithm where weupdatetheweightsin
order to obtain the predicted output
Here, x1,x2,…,xn (independent variable)arethefeaturesofthe
data and b0,b1,b2,…,bn are the weights that are updated and y
is the output(dependent variable)
B. K-nearest neighbor
It is a supervised learning algorithm where value of k is
predefined and changes in the value of k will give rise to
variations in the performancemetrics.Inthisalgorithm,data
points are plotted according to their independent features
and then based on the value of k, a radius is formed around
the datapoint which is to be classified so as to include k
nearest datapoints around it. Value of k has to be an odd
number for classification into two groups.
C. Naive Bayes
Naive Bayes is probabilistic classifier algorithm which
make use of Bayes theorem of probability or Bayes rules.
Naive Bayes algorithm is called naïve because it makes
assumption that the occurrence of certain feature is
independent of occurrence of other features and it is called
Bayes since it uses Bayes rule or law for the classification.
This algorithm predicts membership probabilities for each
class such as the probability that given record or data point
belongs to a particular class. The class with the highest
probability is considered as the most likely class.Thisisalso
known as Maximum A Posteriori (MAP).
Mathematically,
where,
P(A) is the probability of hypothesis A being true. This is
known as the prior probability.
P(B) is the probability of the evidence (regardless of
the hypothesis).
P(B|A) is the probability of the evidence given that
hypothesis is true, which is called Posterior.
P(A|B) is the probability of the hypothesis given that
the evidence is there, which is called likelihood.
D. Random Forest
Random forest is supervised classification as well as
regression algorithm. This algorithm creates a number of
decision tree using the rule-based system. As the name
suggests it forms a forest that is more the number of trees
better is its performance measure. This algorithm first
creates a bootstrapped data from the original data sets.
Then, it creates a decision tree on the basis of the
bootstrapped data. We can create ‘n’ number of divisions of
feature either by selecting one feature, two or n features.
This divided feature data is then converted to decision tree
and these trees are parsed for the data to be classified and a
decision is arrived on the basis of maximum votes for the
class
E. Support Vector Machine
SVM is a supervised algorithm. In this algorithm the
datapoints are separated into a class with the hyperplane
which is as wide as possible from every class. This
hyperplane is constructed with the help of margins of every
class which is created from support vectors. Hyperplane is
constructed using kernel trick which describes whether the
problem stated is nonlinear classifierorlinearclassifier.SVM
algorithm gives good results with even largedatasetsaswell
as large feature space.
Hyperplane in the case of binary SVM classifier has the
equation as:
If
If
where is normal vector to the hyperplane, θi denotes
classes & xi denotes features. The Distance between two
hyperplanes is , to maximize this distance denominator
value should be minimized i.e, should be minimized.
3. Methodology
The main aim of this system is to predict the verdict of a
judicial case based on the verdict of previous Indian judicial
cases. Various prediction model is introduced in previous
section with their fair shares of pros and cons and thus, we
strive to conduct the performance analysis of thementioned
algorithms. Feature extraction for the Indian judicial case is
carried out by textual analysis of the case description in the
prepared case database.
3.1 Datasets and Features
A wide range of data is collected for this system from the
various legal Indian judicial sites (ref). This collected data is
then prepared according to the need of the algorithms for
the training and testing. Prepared data consists of corporate
case description with their judgement in the corresponding
column. This data prepared is then fed for splitting into
training and testing set. We split our prepared data in the
training and testing set of particular size ranging from 80%-
20% to 70%-30%
To apply the prediction models on the data we have to
described its feature. This step of feature extraction is
performed by using the textual information of the cases and
removes the stopping words or words irrelevant todescribe
the feature of the case, thus we use vectorization technique
to define the weightage of a word in the description of the

case. Apart from giving this information, vectorization says
how important that word is to that documentwith respectto
the corpus.
Vectorization technique is also used to convert our
feature extracted into a numerical value which then can be
used for the evaluation of the models and can help us to plot
the point graphically.
3.2 Performance metrics
We have various measure to analyze the performance
measures of all five algorithm. Before we dive into the
performance metrics term we need to know about true
positive, true negative, false negative and false positive
value.
True positive(tp): values which are correctly predicted
positive value
True negative(tn): values which are correctly predicted
negative value
False positive(fp): when values predicted are positive but
target Is to be negative
False negative(fn): when values predicted are negative but
target is to be positive
a. Accuracy:
It is one of the measures for the performance analysis and
it is the ratio of correctly predicted results to the total actual
result of the dataset.
Accuracy=
b. Precision:
It is the ratio of correct predicted positive result to the
total positive predicted result.
Precision=
c. Recall or sensitivity
It is the ratio of correct predicted positive result to total
result in that class
Recall=
d. F1 score:
F1 score is used when their uneven class distribution
feature where accuracy is not as useful as f1 score.it is the
weighted average of precision and recall.
F1score=
3.3 Parameters
We have implemented all the five-predictionalgorithm in
python using scikit library of the python. We have evaluated
performance analysis of all the algorithm on the basis of
80% of training dataset and 20% of testing dataset. Various
parameters involved in the algorithm are described here.
In SVM we use kernel trick as linear(hyperplane) for the
classification of the data into winning and losing
In random forest we create 20 number of decision trees
for the classification of the dataset.as the number of trees
increase performance increase but the complexity and
processing time also increases.
In KNN, we have to define the value for the k which is
used to form the radius for its decision to predict the output
hence we used k=5 as the parameter for KNN.
In Naïve Bayes, we have three types of prediction model,
i.e., gaussian, multinomial and Bernoulli. We have used
multinomial type since it uses discrete counts and used in
text classification which is our base of the judicial prediction
system.
4. Experimentation and Results
In order to gain the maximum accuracy of prediction, the
above predictive models are tested with the same data set.
The accuracy of each predictive model is noted and the
prediction model giving maximum accuracy will be
implemented in the proposed system. The Fig-1 typically
represents the report that is generated after testing is
performed on each prediction or classification algorithm.
Fig -1: Performance of SVM.
Thus, a classification report will be generated for every
algorithm. This classification reportcanbeobtainedbyusing
classified in python. The below is the classification result of
SVM algorithm. It is very crucial to understand the
terminologies used in classification report, they are
explained in the above section.
The classification report is generated for Naive Bayes,
Logistic Regression, Random Forest, k-nearest and SVM
algorithms. While Fig1 gives the classification report for
SVM. Fig-2, Fig-3, Fig-4 and Fig-5 represent Naive Bayes,
Logistic, Random Forest k-nearest respectively.
Fig -2: Performance of Naive Bayes.

Fig -3: Performance of Logistic Regression.
Fig -4: Performance of Random Forest.
Fig -5: Performance of K-nearest neighbor.
After studying each predictive model and determining their
respective accuracy, the precision value of each predictive
model is plotted against each other.
Thus, it can be clearly interpreted that the predictions of
SVM algorithm is more accurate than the rest predictive
model.
Table -1: Precision of Prediction Models for chosen
datasets
Prediction Models
Precision
0 1
Naïve Bayes 0.80 0.83
Logistic Regression 0.67 0.62
Random
Forest
0.33 0.50
K-nearest 0.67 0.80
Support Vector
Machine
0.83 1.00
6. CONCLUSIONS
The advent of Artificial Intelligence has remarkably
changed the perspective towards the applications of
computational power in day to day life. With the robust
collaboration of Statistics, Computer Science and
Engineering, the machines can now adapt to the changing
environment and easily optimize the problems that were
previously non-computational.
The frequent developments in the field of data mining
and Big Data has further given impetus to the applications of
computational power to more diverse fields. Thus, the ever-
evolving technology has led to prodigious innovationsinthe
society. This System has been largely inspired by the recent
developments made in technology and aims at trussing the
gap between Man and the Legal System. In present-day, the
field of law has touched every aspect of
human living. The field of law largely regulates the
philosophy of Life. This has caused frequent acquittance of
citizens with the judicial system. Thus, the dependence on
the legal systems is tremendous. Also, due to the rapid
change in the ideology of living and expeditious evolution,
the judicial system is facing immense pressure. All this have
largely penalized the civilians in terms of renumeration and
time.
This system is with a strong motive to provide legal
assistance to the civilians by predicting outcome of the case
that the citizen will approach the legal system.Thissystemis
estimated to provide the prediction of outcome win or loss
with an accuracy of 78%. Thus, the system will oblige,
encourage civilians in making credible decisions with their
legal proceedings and outreach the citizens with the
legislature for the betterment of the society

REFERENCES
[1] Stefaine Brninghaus and Kevin D. Ashley Predicting
Outcomes of Case- based Legal Arguments, Learning
Research and Development Center and Intelligent
System Program, School of LawUniversityofPittsburgh,
PA
[2] Daniel Martin Katz ,Michael J. Bommarito II , Josh
Blackman A general Approach for predicting the
behaviour of Supreme Court of the United States, April
12,2017..
[3] Zhenyu Liu , Huanhuan Chen A Predictive Performance
Comparison of Machine Learning Models for Judicial
Cases, IEEE,2017.
[4] Octavia-Maria ulea , Marcos Zampieri, Mihaela Vela,
Josef van Genabith PredictingtheLawAreaandDecisions
of French Supreme Court Cases,2017.
[5] Derek Yan ,Tiyangi Wang , Patrick Chase Personal Legal
Counselor and Interpreter of Law via Machine Learning,
FINAL REPORT FOR CS229,FALL 2014.
[6] Sushanta Kumar,P. Krishna Reddy, V. Balakista
Reddy,AdityaSingh Similarity Analysis of Legal
Judgments,ACM,2011.
[7] Top 20 Landmark Judgements ofCorporateLaw.(2019,
Feb 05) Retrieved from https://lawnn.com/top-20-
landmark-judgements-
[8] 40 Latest Corporate Law Cases Law in Simple Format.
(2019,Feb06) Retrieved from
https://www.caclubindia.com/forum/40-latest-
corporate-laws-case-laws-in-simple-format-124405.asp
[9] Case laws on company law. (2019, Feb 10). Retrieved
fromhttps://indiankanoon.org/search/?formInput=case
%20laws%20on%20company%20law
[10] Top 25 Cases of Tax & Corporate Laws Reported in
2017. (2019,Feb15) Retrieved from
https://www.taxmann.com/blogpost/2000000282/top
-25-cases-of-tax-corporate-laws-reported-in-the-year-
2017.aspx
[11] CaseMine: The Most Granular Mapping of US Case
Law.(2019,Feb 15) Retrieved from
https://www.casemine.com/home.
[12] Kanoon(2019,Feb17) Retrieved from
https://indiankanoon.org/
BIOGRAPHIES
Prof. Priyanka Bhilare
Asst. Professor in Computer
Engineering Department
Rajiv Gandhi Institute Of
Technology,
Mumbai.
Ms. Neha Parab
Pursuing B.E. in Computer,
Technology,
Mumbai.
Ms. Namrata Soni
Technology,
Mumbai
Ms. Bhakti Thakur
Technology,
Mumbai

IRJET- Predicting Outcome of Judicial Cases and Analysis using Machine Learning

More Related Content

What's hot (20)

Similar to IRJET- Predicting Outcome of Judicial Cases and Analysis using Machine Learning (20)

More from IRJET Journal (20)

Recently uploaded (20)

IRJET- Predicting Outcome of Judicial Cases and Analysis using Machine Learning