IIIT Hyderabad Study Takes A Data-Driven Look At Fairness In Justice Delivery System

Subjectivity is an inherent part of decision-making whether it’s a doctor’s diagnosis, a manager’s approval, or even a judge’s ruling. All of these are shaped by experience, context, workload, and even subtle cognitive bias. While such variation is human, it can also lead to troubling inconsistencies – especially in systems meant to be fair and objective. In a research paper titled, “Data Cube for Exploring Anomalies in Justice Delivery: An Experiment on Indian Judgements”, IIIT-H researchers Sriharshitha Bondugula, Prof. Krishna Reddy P, and Narendra Babu Unnam in collaboration with Prof. Santhy KVK, NALSAR, have shown that legal systems are no exception to discrepancies and proposed a data analytics based framework to identify disparities/anomalies in court judgements.

According to the researchers, judges, like all decision-makers, bring with them “their unique interplay of personal experiences and cognitive processes, “which makes some degree of variation in judicial decisions inevitable. Decisions such as custody arrangements, bail grants, and sentence imposition, are often subject to the judge’s discretion, the authors note, adding that this can “result in disparities for similar cases across different individuals and courts.” Such unequal treatment for similar cases can have consequences affecting liberty, justice and public trust.

Identifying Anomalies

Rather than focusing on individual intent or motivation to answer the question of how one can systematically identify anomalies and disparities in judicial decisions, the research proposes a framework to identify patterns – using data to highlight where decisions diverge unexpectedly. This approach mirrors practices already used in medicine and marketing domains. The researchers argue that similar analytical approach can be applied to the legal domain, helping institutions detect where decisions may be inconsistent or unfair.

The Wide Space for Interpretation

Indian criminal law clearly defines what punishments are permissible but leaves wide room for interpretation. Under Section 53 of the Indian Penal Code (now replaced by the Bharatiya Nyaya Sanhita), punishments range from fines to death sentences. While minimum and maximum penalties are specified, the gap between them is often large. “The gap leaves room for judges to interpret and quantify the severity of the crime based on contextual nuances,” the paper explains. Judges routinely consider subjective factors: the accused’s background, prior criminal history, intent, brutality of the act, the victim’s age, and mitigating or aggravating circumstances. What counts as serious in one courtroom, however, may be viewed differently in another.

What They Did

To study these differences systematically, the researchers turned to Online Analytical Processing (OLAP) – a technique widely used in sales, healthcare, and marketing to analyse large, multi-dimensional datasets. “However, so far, no effort has been made to extend the data cube-based framework to explore anomalies in the legal domain,” state the researchers. Using the proposed data cube based approach, the researchers analysed sentencing patterns across multiple dimensions at once – crime type, state, year, punishment type, prison duration, and fine amount – enabling them to zoom in and out of the data to spot trends and outliers.

Turning Judgments into Structured Data

One of the biggest challenges was of extracting data since court judgments are not written as neat datasets. The IIIT-team used manually extracted data from nearly 3,500 Indian criminal cases involving murder, kidnapping and rape, covering judgments between 2005 and 2010. They then used large language models (LLMs) to convert messy, unstructured legal text into structured variables. This allowed them to build a clean dataset capturing verdicts, punishment types, prison durations, and fines – the foundation for their analytical framework.

What The Data Revealed

Once the data cube was built, striking patterns emerged. Across states, the study found clear variations in fines and prison terms for similar crimes. For example, in homicide cases, fines were rarely imposed across most states – except Kerala, which showed significantly higher fine amounts and greater variation. In aggravated rape cases, Himachal Pradesh imposed higher fines compared to Delhi, even when average values were similar elsewhere. For aggravated kidnapping, longer prison terms were observed in Rajasthan than Haryana. “These patterns warranted deeper exploration of this subspace,” states Sriharshitha, co-author of the paper, emphasising that such differences are not obvious without multi-dimensional analysis.

Discretion Needed, But Consistency Matters

According to the researchers, while discretion is necessary for individualised justice, ensuring a balance between flexibility and consistency fosters fairness. They cite other countries which have grappled with this issue before, like the United States where despite the introduction of sentencing guidelines in the late 1980s to temporarily reduce subjectivity and improve uniformity, debates around rigidity versus fairness continue.

It is hoped that the research will encourage the investigation of a comprehensive data cube-based framework to explore disparities in the legal domain for India and abroad. Periodically, such a system will identify disparities and foster corrective measures. The researchers have also made their dataset – the Indian Judgements Punishment Data (IJPD) – publicly available. Future plans include extracting data directly from full judgment texts and incorporating additional dimensions such as the age of the accused, the victim’s age, and even the judge’s gender.

In a justice system built on human judgment, the study offers a powerful reminder that while data cannot replace discretion, it can illuminate where discretion begins to look like disparity.