Volume 8, Issue 2, February – 2023 International Journal of Innovative Science and Research Technology

ISSN No:-2456-2165

Indian Penal Code Recognition Using Multiclass

Classification Algorithms in Machine Learning
Prakash Aryan
Department of Computer Science and Engineering
Purnima University
Jaipur, Rajasthan

Abstract:- As of May 2022, over 4.7 crore cases are pending With the digitalization of the Indian Judiciary the process
in courts across different levels of the judiciary. This has of judgment can be enhanced and greatly reduced. The Indian
resulted in a ponderous burden on the Indian judiciary Penal Code is the official penal code of the Republic of India.
system. According to Experts the reason why we have this It is a complete code designed to cover all aspects of criminal
problem is due to lack of infrastructure in the Indian law. The draft of the Indian Penal Code was prepared by the
Judiciary System. With this study I build a machine First Judicial Council, chaired by Thomas Babington Macaulay.
learning model that can determine the applicable Indian The draft was based on a simple codification of English law,
Penal Code according to the description provided by the but at the same time borrowed elements from the Napoleonic
user. I have used two machine learning approaches, one vs Code and the Louisiana Civil Code of 1825.The IPC has been
rest and one vs one multi-class classification. The one vs rest successful by and large in its attempt to prosecute and punish
classification using Logistic Regression has a 99% accuracy individuals who commit the crimes that are defined in this
on training data. Code, but like Sedition there have been certain other provisions
that have invited scrutiny time and again. IPC as a statute has
Keywords:- Machine Learning, Natural Language Processing, survived and flourished over the last 160 years, which speaks
Multiclass Classification. volumes about its effectiveness as a penal code of high stature.
However, over the years he has failed to get rid of some
I. INTRODUCTION positions that smell of colonialism, for example. Rebellion. The
Malimat Commission's report advocating criminal justice
Between 2015 and 2019, the average number of pending reform provided an opportunity for Congress to update the
cases was approximately 1.8 million per year, but the High Code and other criminal laws. Seventeen years have passed
Court currently lists 5.8 million pending cases. In most cases, since the report was submitted, but no concrete action has been
the number of cases closed is lower than the number of cases taken in this regard. It's time for the legislature to intervene to
opened, which continues to exacerbate the problem.[1] This has make the code more modern than the British colonies. High
resulted in a huge backlog of cases and has in turn resulted in courts stepping in and breaking the law does not reflect well on
delay of justice. Experts have looked into this problem and the legislature. Because that's the legislature's job in the first
determined the various reasons as to why this is happening. The place.[3]
foremost reason is the pending vacancies in the judiciary sector.
One of the ways we can solve this is by encouraging digital In this study, I aim to reduce the time taken on determining
courts in our country. Raj said there is a need to "expand and the applicable Indian Penal Code and the confusion around it.
strengthen the system of digital courts, i.e. online courts, so that For a common man finding the appropriate section that is
majority of the cases can be decided through the online system, applicable in a given situation is very complicated and the
which will popularize the judiciary, decentralize it and expedite person will have to seek a legal consultation which costs a lot
the process of adjudication, across the country."[2] both financially and timely. With this model we can predict the
applicable IPCs in a particular situation based on the description
Justice has often been delayed for people even for the provided by the user. This can help an individual get an idea of
fundamental rights. A lot of farmers have been affected due to the seriousness of the situation and where do they stand in the
this slump in our judiciary system. People have to be released entirety. The problem involves classifying the correct IPC
after 20 to 30 years because enough evidence could not be applicable based on the user input. For this we will be using two
gathered to acquit them of charges. Then we come across cases of the machine learning algorithms which are multi class
where people have waited years and even passed away after classification algorithm. One vs Rest and One vs One is used
which their descendants had to fight the cases on their behalf and the accuracy for both of these models are compared on the
and they were granted justice after 20 years. Such incidence training set.
puts the trust of people on democracy on a thin line and people
lose hope.

IJISRT23FEB1360 www.ijisrt.com 1925

Volume 8, Issue 2, February – 2023 International Journal of Innovative Science and Research Technology
ISSN No:-2456-2165
II. RELATED WORK presentation of knowledge—lawyers are grappling with their
professional obligations, ethical duties, and what it means for
The use of machine learning in the field of legislation has the future of legal practice. Through in-depth, semi-structured
been on the uprise. This is because it has made the process of interviews of experts in the e-discovery technology space—the
legislation very simple for both the working professionals technology company representatives who develop and sell such
(lawyers and judges) and for the general public. Application of systems to law firms and the legal professionals who decide
artificial intelligence in the legal world various areas of AI in whether and how to use them in practice—we shed light on the
the legal sector can be discovered with adequate research and organizational structures, professional rules and norms, and
careful comprehension of the legal industry by IoT app technical system properties that are shaping and being reshaped
development companies. The current Artificial Intelligence by predictive coding systems. [5]
applications in the industry can be categorized into six main
parts: In February 2018, an artificial intelligence system
developed by Israeli legal technology firm lawgeex won more
 Due Diligence: than 20 leading lawyers with rich experience in a standard
Lawyers use Artificial Intelligence tools to perform due business contract review competition. In addition, the average
diligence and uncover background information. In light of the accuracy is more than 9% higher than that of humans, and in
current scenario, developers have opted to integrate a slew of the whole process, the system that needs to complete tasks that
new features, including agreement review, legal inquiry, and humans need to complete in 92 minutes or less completes in 26
electronic media for this section of the industry. seconds, so speed and efficiency, he has committed "double
murder" of people. [6]
 Prognostication Technology:
Artificial Intelligence (AI) aids in the generation of III. MATERIALS AND METHODS
outcomes for legal investigations and agreement evaluations.
This characteristic of AI programming appears to be extremely A detailed overview of how the project is carried out has
beneficial to legal firms and industries. been discussed in this section. Various important parts of
building a machine learning model such as Data
 Legal Mechanism: acquisition, Data preprocessing, feature selections and
Lawyers can use artificial intelligence technology to importance are also described.
obtain information from previous or past instances. This data
can also be used to track referee instructions and predictions.  Software Requirements
This technology will become increasingly important worldwide  Operating System -Pop!_OS, Release –21.04
in the near future.  Python 3.8.3
 Jupyter NotebookVersion –6.0.3
 Documentation Mechanisms:
Various types of software are used in the legal industry to  Hardware Requirements
develop documents that support the collection of data and  CPU –Ryzen 5 3550H Quad core
information. There are many useful documents in the field of  RAM –8gb ddr4
law firms.  Storage-20gb
As a result, it is really profitable.  Dataset
We have used the Indian Penal Code dataset. To train our
 Intellectual Property: model better we have used 200 selected sections so that we can
AI algorithms show lawyers how to sift through massive first test our model according to which algorithms suits it the
IP files and extract meaning from a multitude of powerful texts. most. The sections are up to 301. [7]
 Electronic Receipts:  Algorithms
For a long time, lawyers created their own receipts. After Machine Learning (ML) is the scientific study of
these companies adopted AI software development technology, algorithms and statistical models that computer systems use to
lawyers' accounts went electronic.[4] perform a specific task without being explicitly programmed.
Learning algorithms in numerous operations that’s we make use
These systems and the companies offering them are of diurnal. Every time a web hunt machine like Google is used
reshaping relationships between lawyers and clients, to search the internet, one of the reasons that work so well is
introducing new kinds of professionals into legal practice, because a literacy algorithm that has learned how to rank web
altering the discovery process, and shaping how lawyers runners. These algorithms are used for colorful purposes like
construct knowledge about their cases and professional data mining, image processing, prophetic analytics, etc. to
obligations. In the midst of these shifting relationships—and the name a many. The main advantage of using machine learning is
ways in which these systems are shaping the construction and that, once an algorithm learns what to do with data, it can do its

IJISRT23FEB1360 www.ijisrt.com 1926

Volume 8, Issue 2, February – 2023 International Journal of Innovative Science and Research Technology
ISSN No:-2456-2165
work automatically. In this paper, a brief review and unborn Accuracy of a model is one of the most used evaluation
prospect of the vast operations of machine learning algorithms technique. It tells us how accurate our model is in making
has been made.[8] predictions. Accuracy is calculated by – Correct predictions/
Total predictions.
One vs. all provides a way to work double bracket. Given
a bracket problem with N possible results, a one-vs. - All result TP+TN
consists of N separate double classifiers — one double classifier Accuracy =
for each possible outgrowth. During training, the model runs 𝑇𝑃 + 𝑇𝑁 + 𝐹𝑃 + 𝐹𝑁
through a sequence of double classifiers, training each to
answer a separate bracket question [9]

Some classification algorithms permit the use of further

than two classes by design. Others circumscribe the possible
issues to one of two values (a binary, or two- class model). But
indeed double bracket algorithms can be acclimated for multi-
class bracket tasks through a variety of strategies. This element
implements the one- versus- bone System, in which a double
model is created per class brace. At vaticinator time, the class
which entered the most votes is named. Since it requires to fit
n_classes *(n_classes- 1)/ 2 classifiers, this system is generally
slower than one- versus- all, due to its O (n_classes2)
complexity. Still, this system may be profitable for algorithms
similar as kernel algorithms which do n’t gauge well
withn_samples. This is because each individual literacy
problem only involves a small subset of the data whereas, with
one- versus- all, the complete dataset is used n_classes
times.[10]

Logistic Regression is a classification algorithm that

performs well for linearly separable classes. It is designed for
two class classification but with the help of One-vs-Rest (OvR)
classifier it can handle multi-class problem also. The output of
logistic regression is binary dichotomous.[11]

Support Vector Machines are classification algorithms

that are based on determining vectors that in turn creates a
hyperplane. The classes are divided based on this hyperplane.
[12]

Natural Language Processing - Natural languages are

those languages that are spoken by the people. Natural language
processing cinctures everything a computer needs to understand
natural language and also generates natural language. Natural
Language Processing is a subfield of Artificial Intelligence and
verbal, devoted to make computers understand the statements
or words written in mortal languages. A Natural language also
known as ordinary language that's spoken or written by people
(humans) for general purpose communication. Natural
language came into actuality because when stoner wishes to
communicate with the computer we can’t force the druggies
to learn machine specific language so this principally caters to
directors or children’s who don't have enough time to learn new
specific languages or get professed in them. Languages can be Fig – 1: Flowchart of the Model Implementation
any like Hindi, French, English, and Chinese etc. A language is
a system, a set of rules or set of symbols. [13]

IJISRT23FEB1360 www.ijisrt.com 1927

Volume 8, Issue 2, February – 2023 International Journal of Innovative Science and Research Technology
ISSN No:-2456-2165
IV. RESULTS [10]. “One-vs-One Multiclass - Azure Machine Learning |
Microsoft Learn⁤.” https://learn.microsoft.com/en-
We have used accuracy parameter to check performance us/azure/machine-learning/component-reference/one-vs-
of our model. This is done on the training set of our dataset. We one-multiclass (accessed Feb. 28, 2023).
observed that using one vs one classification using support [11]. D. W. Hosmer and S. Lemeshow, “Applied Logistic
vector machines gave us an accuracy score of 97% and when Regression,” Applied Logistic Regression, Jan. 2005, doi:
we used one vs rest with logistic regression we got an accuracy 10.1002/0471722146.
score of 93%. This is due to the fact that each data point on our [12]. A. J. Smola and B. Schölkopf, “A tutorial on support
dataset is comparable to each and every other dataset and cannot vector regression,” Stat Comput, vol. 14, no. 3, pp. 199–
be grouped together. We had to vectorize our words and 222, Aug. 2004, doi:
sentences into arrays because machine learning needs 10.1023/B:STCO.0000035301.49549.88.
numerical data for processing. [13]. “https://citeseerx.ist.psu.edu/document?repid=rep1&type
=pdf&doi=eeace1d14e266a5cd44fe781a874c662928602f
V. CONCLUSION AND DISCUSSION d⁤.”
https://citeseerx.ist.psu.edu/document?repid=rep1&type=
With this model we saw that using the correct pdf&doi=eeace1d14e266a5cd44fe781a874c662928602fd
methodology and the right dataset can aid the general public (accessed Feb. 27, 2023).
and the people working in legislation. The dataset which we
have is not very sophisticated and thus we could not check
accuracy on trainset. This is one of the potential we can work Author –
on. We can include the entire dataset of the Indian Penal Code This Paper is submitted by Mr. Prakash Aryan. He is an
instead of selective sections. We can also include more undergraduate student pursuing his course in computer science
description of each dataset so that our model does not underfit. and engineering at Poornima University Jaipur. His interest lies
in Machine Learning, Natural Language Processing and
REFERENCES Computer Vision.

[1]. “How to Start Resolving the Indian Judiciary’s Long-

Running Case Backlog - Carnegie Endowment for
International Peace.”
https://carnegieendowment.org/2021/09/09/how-to-start-
resolving-indian-judiciary-s-long-running-case-backlog-
pub-85296 (accessed Feb. 24, 2023).
[2]. “India’s huge backlog of court cases delays justice:
Experts⁤.” https://www.aa.com.tr/en/asia-pacific/indias-
huge-backlog-of-court-cases-delays-justice-
experts/2638784 (accessed Feb. 24, 2023).
[3]. “Indian Penal Code (IPC) - History, Structure & Recent
Developments⁤.” https://byjus.com/free-ias-
prep/indian-penal-code/ (accessed Feb. 27, 2023).
[4]. “Role of artificial intelligence in law - iPleaders.”
https://blog.ipleaders.in/role-of-artificial-intelligence-in-
law/ (accessed Feb. 27, 2023).
[5]. “AUTOMATED DECISION SUPPORT
TECHNOLOGIES AND THE LEGAL PROFESSION on
JSTOR.” https://www.jstor.org/stable/26954398
(accessed Feb. 27, 2023).
[6]. T. Hu and H. Lu, “Study on the Influence of Artificial
Intelligence on Legal Profession,” Jan. 2020, doi:
10.2991/AEBMR.K.191225.184.
[7]. “THE INDIAN PENAL CODE”.
[8]. “Machine Learning Algorithms-A Review,” 2019, doi:
10.21275/ART20203995.
[9]. “Multi-Class Neural Networks: One vs. All | Machine
Learning | Google Developers.”
https://developers.google.com/machine-learning/crash-
course/multi-class-neural-networks/one-vs-all (accessed
Feb. 27, 2023).

IJISRT23FEB1360 www.ijisrt.com 1928