Reducing Arbitrariness in Criminal Investigation

A complaint typically signifies the commencement of criminal proceedings. In Mainland China, when the People’s Court, the People’s Procurator or public security authorities examine the materials related to reports, accusations and surrenders, they must determine if there are criminal facts warranting legal action (Zhang Reference Zhang2023). If such facts are identified, a lawsuit must be initiated, and evidence substantiating the occurrence of a crime is necessary for an arrest to take place. In the US legal framework, the term “probable cause” is commonly employed to justify searches, seizures or arrests (Smiley-McDonald et al. Reference Smiley-McDonald, Attaway, Wenger, Greenwell, Lambdin and Kral2023). However, the determination of whether an investigative agency believes or possesses probable cause inherently involves a subjective element that is an inescapable aspect of investigative processes. In this context, some scholars refer to the police culture of suspicion as sixth-sense suspicion, positing that this intangible sense of suspicion reflects the true capabilities of law enforcement, as recognized by their peers (Mydlowski and Turner-Moore Reference Mydlowski and Turner-Moore2025).

Investigators must take into account a variety of factors during the course of their inquiries when determining the appropriateness and method of implementing specific coercive measures. For instance, in Mainland China, public security agencies are required to thoroughly evaluate the nature of the case, the surrounding circumstances, the potential social repercussions and the suspect’s social connections, among other considerations (Evans, Leese, and Rychnovská Reference Evans, Leese and Rychnovská2021). These factors influence the selection of the specific location, individual and scope of the investigative activities. In the context of American criminal justice, investigators assess numerous elements such as socio-economic status, race, age, attire and gender to ascertain whether an individual should be regarded as a suspect. In an age characterized by limited data, decision-making by investigators largely relies on their training, personal experiences and intuitive judgment. Their choices are often influenced by the primacy effect, which encompasses biases, investigative inertia and cognitive limitations, alongside other unpredictable factors. Whether pursuing a criminal suspect or implementing measures against one, investigators face the challenge of navigating the intricate realities of each case while adhering to pressing time constraints, which inevitably constrains their decision-making capabilities.

In the age of big data (Jiang Reference Jiang2023), applications leveraging this vast information can significantly enhance the precision of decision-making processes. In contrast to the random and often chaotic choices made by human cognition, big data enables researchers to receive timely and relevant guidance through established algorithms and authoritative sources. The foundation of these big data algorithms lies in statistics, which is regarded as a convergence of mathematical reasoning, statistical analysis, computer science and domain-specific knowledge. The predetermined nature of these algorithms helps to minimize the subjectivity and biases that individual researchers might possess. The shift from the small data era to the big data era is poised to improve both the efficiency and fairness of criminal investigations.

Furthermore, leveraging big data can enhance the precision of criminal investigations and mitigate tensions between law enforcement and the community (Akinbowale, Mashigo, and Zerihun Reference Akinbowale, Mashigo and Zerihun2023). In the context of small data, investigators often require direct engagement with the public to identify and filter potential suspects, particularly at transportation hubs such as railway and subway entrances. When the public collaborates with investigators during specific operations, they are typically required to present identification. In this scenario, despite public cooperation and comprehension of investigative procedures, there remains a risk of state intrusion into citizens’ private lives, potentially fostering a psychological divide between the populace and governmental entities. Conversely, in the realm of big data, insights could be derived from the big intelligence data platform. Through the application of big data analytics, investigators can pre-emptively profile individuals involved and anticipate potential routes, allowing for targeted interventions that reduce friction with the public.

From Post-Event Investigation to Pre-Event Prevention

Prior to the implementation of big data analytics, the volume of data utilized was minimal. Criminal investigations typically adhered to a sequence comprising incident occurrence, followed by investigative intervention, and concluding with evidence collection. This process inherently involves a time delay between the actions of investigative agencies and the actual commission of a crime. Consequently, investigations tend to be reactive, focusing on remedial and punitive measures rather than adopting a proactive approach. A significant advantage of big data lies in its predictive capabilities. By leveraging big data, investigative agencies can strategically allocate their resources based on predictive analytics, thereby mitigating the incidence of crimes and ensuring timely evidence collection when related offences transpire.

An essential application of big data in crime prevention is the development of crime maps (Roy and Chowdhury Reference Roy and Chowdhury2023). Crime mapping involves the integration of existing crime incident and arrest report data into a computer system, where it is processed and visually represented on a map. This visual–spatial representation enhances the utility of the data. By analysing hotspot maps derived from historical crime data, law enforcement officials can accurately identify patterns, such as the operational areas of criminal suspects and the current prevalence of specific crimes (Dau et al. Reference Dau, Dewinter, Witlox, Beken and Vandeviver2023). Evidence indicates that the implementation of crime maps has led to a half reduction in shooting incidents and about a three times increase in the number of weapons confiscated (Mazeika Reference Mazeika2023).

The development of crime mapping and the efficacy of proactive measures by investigators are underpinned by various criminological and sociological theories, among others. For instance, routine activity theory posits that three conditions must converge for criminal activity to take place (Okafor Reference Okafor2023): the presence of a potential offender; the availability of suitable targets; and the absence of capable guardianship (Vu Reference Vu2023). When these three factors coincide in a specific time and location, an opportunity for crime emerges. Furthermore, situational crime prevention theory suggests that criminal behaviour stems from the rational decisions made by offenders. Criminals often subconsciously weigh multiple factors when contemplating the commission of an offence. Consequently, it is imperative for investigators to enhance the perceived risks associated with criminal actions while simultaneously diminishing the potential rewards.

The approach to criminal investigation has evolved significantly with the advent of big data. It has transitioned from a traditional passive model to a proactive framework characterized by crime prediction, followed by investigation intervention, and ultimately leading to crime prevention. This new investigative paradigm not only addresses the deficiencies in supervisory resources but also enhances the perceived risks associated with criminal activities, thereby creating an external deterrent against such behaviours.

Diverse Viewpoints to Assist in the Identification of Cases

Data mining involves the automated examination of extensive datasets through specialized computer algorithms, aimed at uncovering concealed relationships, patterns and trends within the data. This process ultimately yields new insights for decision-makers. By processing and interpreting existing data, big data techniques can be utilized to extract a diverse range of information.

Protection of Rights during Data Collection and the Effect

The information sources for investigative agencies are primarily derived from two main avenues. The first source pertains to the daily operational activities of these agencies, while the second is linked to the data provided by network operators. Many nations explicitly emphasize the need to expedite the development of a public security video surveillance system from a public safety and social security standpoint. It advocates for a strategic approach that prioritizes the construction, integration and application of public security video surveillance (Şengönül et al. Reference Şengönül, Samet, Abu Al-Haija, Alqahtani, Alturki and Alsulami2023), aiming to enhance both the coverage density and the quality of surveillance systems in public areas. In recent years, the comprehensive utilization service platform for public security big data has facilitated the collection of extensive GPS and personal identification information, which has been effectively utilized by investigative agencies and other entities in their operations. The integration of pervasive electronic surveillance with biometric data and facial recognition technology enables the identification of individuals in public or semi-public spaces. In the current information era, data holds paramount importance. As the quality of information increasingly underpins power dynamics, the complex interplay between state authority and civil liberties reveals a strained yet interdependent relationship.

In contrast to traditional societies, the dynamics between the state and its citizens in the digital age involve new participants, including network operators and social media platforms, which collect substantial amounts of personal data from individuals in the course of their operations. Within the realm of criminal justice, these Internet operators are typically obligated to comply with requests for information from law enforcement agencies. This inherent bias among Internet operators necessitates that ordinary citizens remain cautious about third parties accessing their personal data, particularly considering the encroachment of significant governmental authority. Consequently, citizens’ personal information becomes obscured from both the state and network operators, leading to a gradual erosion of privacy rights (Oatley Reference Oatley2022) and the potential for citizens to be rendered entirely transparent. Furthermore, the disparity in information-gathering capabilities between the prosecution and defence exacerbates the imbalance in equality within the judicial process. The extraordinary powers that the government has gained through advancements in technology – such as the ability to penetrate physical barriers, perceive through them, and detect even the faintest sounds – are increasingly alarming, surpassing the threats posed by direct intrusion.

The data collection process is inherently susceptible to bias and superimposition (Ferrara Reference Ferrara2024). As previously noted, areas identified as crime hotspots through the analysis of crime incidents, timings, locations and various other factors tend to draw increased attention from investigators. This creates a logical progression: an abundance of crime data leads to a focus on crime hotspots, which in turn results in greater allocation of investigative resources (Mandalapu et al. Reference Mandalapu, Elluri, Vyas and Roy2023). Consequently, the volume of crimes under investigation is likely to rise. This escalation causes hotspot areas to become even more pronounced, as the number of crimes being investigated and addressed grows. Crime maps generated from the ongoing superimposition of data become ensnared in a self-reinforcing cycle, thereby raising concerns about their objectivity (Kronkvist et al. Reference Kronkvist, Borg, Boldt and Gerell2025).

There are concerns regarding the presence of bias in the data collection process. In certain judicial practices across various countries, the race factor is considered by investigators when assessing the potential threat posed by an individual (Phillips and Bowling Reference Phillips, Bowling and Vogel2020). Consequently, the original database maintained by the investigative agency shows that the incidence of crimes attributed to people of colour significantly surpasses that of crimes attributed to white individuals. The bias inherent in the data collection process, combined with the compounding effects of big data analytics, further increases the likelihood that people of colour will be disproportionately targeted in investigations (Simmons Reference Simmons2021).

The thoroughness of data collection remains uncertain. Research indicates that for certain offences, such as murder and theft, victims typically opt to report incidents to law enforcement. However, in instances of sexual assault (Chan Reference Chan2008), domestic violence and other crimes, only a small fraction of cases are reported. Cases that do not come to the attention of authorities forfeit the opportunity to be documented in official records. This lack of reporting can skew the crime mapping generated through big data, potentially leading to inaccuracies in resource allocation decisions made by law enforcement. For instance, in sexual assault situations, the presence of investigators may induce significant psychological distress, which can deter the perpetrator and prevent further criminal activity. Consequently, the role of big data may be diminished, rendering it an accomplice in certain criminal scenarios. Furthermore, because these specific cases were never officially acknowledged, big data also loses its potential to aid in uncovering such incidents.

Arbitrariness in Data Processing

The objective and scientific nature of big data analysis lends credibility to its data processing techniques; however, it also masks inherent subjectivity and the phenomenon known as the “black box effect” (Hassija et al. Reference Hassija, Chamola, Mahapatra, Singal, Goel, Huang, Scardapane, Spinelli, Mahmud and Hussain2024). This effect refers to a situation where the public can only see the inputs and outputs of a specific device, yet remains unaware of the internal processes that govern information processing within that device. The internal workings of the device are not accessible to the public, thereby creating a black box.

Initially, the establishment of variables is crucial. In the context of analysing a crime map and forecasting criminal behaviour, it is essential for investigators to define relevant variables. The process of identifying and validating these variables inevitably involves human judgment. Typically, when investigators formulate variables for a crime prediction model, they focus primarily on the historical incidence of crimes in the area, while also considering additional elements such as paydays, seasonal variations, hotel placements and possible escape routes. When developing a crime map specifically for theft incidents, the following factors are generally assessed (Kounadi et al. Reference Kounadi, Ristea, Araujo and Leitner2020):

1. (i) frequency of previous theft offences

2. (ii) number of arrests related to property crimes

3. (iii) recent residential patterns

4. (iv) proximity to major thoroughfares

5. (v) geographic distribution of younger demographics

6. (vi) locations of apartments and dining establishments

The selection of specific model variables can differ significantly across various locations and is largely influenced by the investigator’s expertise.

Secondly, the computation of the data itself is crucial. Deep learning represents a fundamental capability of artificial intelligence. Utilizing established algorithms, machines develop new understandings by learning from, summarizing and inferring insights from historical data. Within this framework, the outcomes of machine cognition and inference are expected to demonstrate a progressive enhancement, potentially surpassing human cognitive abilities to some degree (Wu and Wang Reference Wu and Wang2020). Nevertheless, certain researchers have explicitly noted that the deep learning process, which involves extracting features from raw natural signals to accomplish tasks, operates as a black box, lacking interpretability and resembling the basic cognitive functions observed in mammals. Challenges such as the mechanisms by which machines achieve self-learning and the inherent limitations of this process remain significant barriers that current technologies have yet to overcome (Rudin Reference Rudin2022).

The text must encompass a minimum of four components: accuracy, fairness, truthfulness and privacy. Equality signifies that individuals should not face bias or be regarded as offenders based on their personal circumstances. The application of big data in criminal investigations cannot entirely eliminate identity considerations due to the human influence in parameter configuration (Verhelst, Stannat, and Mecacci Reference Verhelst, Stannat and Mecacci2020). Furthermore, the inherent black box effect at the technical level obstructs the achievement of the necessary transparency. Consequently, the integrity of due process is significantly compromised.

Security When Storing Data

For investigative agencies, whether the data is collected based on daily activities or provided by Internet operators, it will eventually be aggregated on the investigative agency’s platform such as a police cloud (Qian et al. Reference Qian, Cheng, Wang, Yang and Li2021). Judging from the police cloud platforms built in various places, they generally integrate local resident information, DNA data information, vehicle identification systems, past criminal case information, etc. By using this kind of police cloud, billons of data could be integrated. Within these massive data, powerful search, comparison and computing capabilities, the police cloud has become a smart investigator and detective of the big data era.

The foundational structure of integrated information systems inherently carries the risk of information leakage. This risk manifests in two primary ways. Firstly, from a technical standpoint, the information available on the Internet is vulnerable to breaches and attacks. Even when a network of information management adheres to the internationally recognized standard, there is no absolute assurance that any specific data will be entirely safeguarded against potential attacks. The inherent characteristics of police cloud systems, such as distributed storage and multi-client access, further increase their susceptibility to such threats (Bhardwaj et al. Reference Bhardwaj, Mangat, Vig, Halder and Conti2021). Secondly, from the viewpoint of investigators, they function merely as users of big data technologies rather than as developers of these technologies. In practice, when public security agencies establish police cloud systems, they frequently pursue collaborative strategies with Internet technology firms. Research indicates that databases containing a high concentration of sensitive personal information are particularly appealing targets for criminals (Oatley Reference Oatley2022), both within and outside the organization. Globally, numerous databases belonging to prominent Internet companies and various government entities have experienced significant breaches.

The economic significance of platforms that aggregate substantial volumes of data, the constraints on objectivity inherent in the present phase of network security technology development, the inability of investigative agencies to manage large-scale data storage, and the public’s misunderstanding regarding the collection, analysis and potential leakage of personal information are all objectively linked to heightened risks in the data storage process.

Improving Data Quality

The foundation for the precision of big data predictive models lies in the dependability of the input data. However, in practice, any influence stemming from mechanical input errors, data bias or insufficient data can render the data incomplete. This issue is particularly evident in the integration of big data within the criminal justice system, including initiatives such as police clouds, intelligent prosecutions and smart courts. The challenge of data quality has consistently been a concern for the academic community, garnering significant attention from numerous researchers (Lauritsen Reference Lauritsen2023). Some scholars have explicitly noted that despite the extensive data collection, the data that is gathered, organized and compiled often fails to meet the standards of legal big data (Janssen et al. Reference Janssen, Brous, Estevez, Barbosa and Janowski2020). In fact, the quality of certain data may fall short of the representative data obtained through sample surveys.

Some examples of police clouds illustrate that the multitude of information sources and the varying formats encountered during the information collection process heighten the risk of errors. Consequently, investigators are required to meticulously review a vast quantity of data individually. Additionally, the notion that compelling individuals to undertake challenging tasks is a misconception; the information available on previous cases is often insufficient, and improvements in the current situation cannot be achieved instantaneously.

It is somewhat awkward to acknowledge that one must depend on advancements in data collection technology to reduce errors stemming from both mechanical and human shortcomings. Simultaneously, one must adhere to the established burden-of-proof principle and apply external pressures to ensure the integrity of the data and enhance its quality. In the context of criminal cases, the responsibility to prove guilt lies with the prosecution, as noted by Peng (Reference Peng2020), except in specific circumstances outlined by law (Chen and Zheng Reference Chen and Zheng2007). When big data analysis serves as the foundation for initiating an investigation, should the defendant challenge the data’s accuracy, the investigator is obligated to disclose the data’s source and verify its correctness. In this scenario, the parties involved can establish the standard of proof based on the mechanism for excluding illegal evidence. This means that the defendant need only present initial doubts, after which the investigator must substantiate the data’s accuracy. If the investigative body fails to demonstrate the information’s accuracy to the preponderance-of-evidence standard, it indicates that the investigation was improperly initiated, warranting a preferential sentence for the defendant based on the circumstances. In China, the reform of the trial-centred litigation system, at least theoretically, aids in enhancing the legitimacy of investigative practices and offers an external institutional assurance for the application of this model.

Restricted Sharing of the Data Processing Procedure

The utilization of big data in crime prediction models remains primarily confined to the implementation of investigative techniques during the preliminary investigation phase. As outlined in current legal provisions (Chen and Zheng Reference Chen and Zheng2007), public security agencies are prohibited from imposing restrictions on measures such as interrogation, questioning, examination, identification, and the collection of evidence and materials, as well as on the personal and property rights of the individual under investigation. Electronic data acquired during the preliminary investigation may serve as admissible evidence. During the investigation phase, it is essential to incorporate a necessary closure aspect to ensure the effective collection and preservation of evidence, thereby preventing its loss.

It is essential to uphold the principle of due process during the trial stage by ensuring public access. Beyond the fundamental requirement of trial openness, it is crucial to incorporate elements such as procedural equality. According to Taniguchi (Reference Taniguchi2000), a key aspect of procedural fairness is that any individual with a vested interest in the outcome of the proceedings, or who may suffer negative consequences from it, has the right to engage in the proceedings and to be affected by them. This includes the opportunity to present their arguments and to counter the claims and evidence put forth by the opposing party. Throughout the trial process, concerns often arise regarding the reasonable grounds established through big data investigations and their resultant judgments. For a rebuttal to the findings of big data investigations and judgments to be valid, it is imperative that the underlying algorithm is disclosed.

In a highly competitive market landscape, it is unlikely that any legal technology firm would reveal the essential components of its algorithm to external parties, much less provide a detailed explanation of its operation in public forums. The pertinent issue is whether the development of a data algorithm qualifies as a trade secret. The fundamental characteristics that define trade secrets include confidentiality, economic benefit, practicality and the implementation of specific protective measures. The inherently confidential nature of big data algorithms, along with the potential economic advantages they can confer to technology firms, aligns perfectly with the established definition of trade secrets.

The commercial confidentiality of the algorithm does not provide sufficient grounds for its non-disclosure (Dijesh, Babu, and Vijayalakshmi Reference Dijesh, Babu and Vijayalakshmi2020). To address the challenge of safeguarding trade secrets while maintaining the transparency of litigation, some scholars have introduced the notion of meaningful transparency. This concept suggests that, in instances where a big data algorithm model is contested in court, the party utilizing the algorithm is not required to reveal the specific model but rather to disclose the primary efforts undertaken by the designer in its development. When both explainable and unexplainable models serve similar functions, the focus should be on verifying the model, which includes identifying influencing factors, constructing decision trees and exploring alternative methodologies. In situations characterized by black box effects, it is advisable to opt for an explanatory model. Current judicial practices indicate that a strategy of limited disclosure effectively balances the need to protect commercial confidentiality with the imperative of ensuring justice in litigation.

Hierarchical Protection of Personal Information

Personal information refers to any data collected through electronic or alternative methods that can identify an individual, either on its own or in combination with other information. This includes details such as a person’s name, date of birth, identification number, biometric data, home address, telephone number and similar information.

In the digital age, the safeguarding of personal information is less robust than it was prior to the advent of the Internet (Dai Reference Dai2020; Shafik et al. Reference Shafik, Kalinaki, Fahim and Adam2025). This vulnerability arises as individuals may inadvertently disclose their information on social media platforms, while certain individuals may deliberately seek to acquire others’ information unlawfully. Current legal frameworks affirm the protection of personal information belonging to individuals.

It is important to highlight that, in contrast to the regulations in the United Kingdom and the United States, the Mainland China’s Criminal Procedure Law (Chen and Zheng Reference Chen and Zheng2007) primarily categorizes the acquisition of citizens’ personal information by investigative agencies as a technical investigation rather than as a coercive measure. The differences in legislative approaches result in varying mechanisms that restrict the exercise of authority. From a definitional standpoint, technical investigation encompasses specialized investigative actions undertaken by public security agencies to gather evidence or apprehend criminal suspects through the use of technical equipment, necessitated by the requirements of criminal investigations and subject to rigorous approval processes. Consequently, due to the systematic classification of investigative activities, the execution of data monitoring by investigators necessitates only internal authorization from the investigative bodies, specifically the approval of the head of the public security agency at the district or municipal level or higher. The role of neutral third-party intervention has diminished. With the reallocation of responsibilities within the procuratorate, its legal oversight function has been further diminished, leading to a significant reduction in the rigidity of supervision. In this context, the safeguarding of citizens’ personal information has entered an increasingly precarious state.

In the realm of criminal justice, it is essential to address the power disparity between investigative agencies and defendants to establish a more equitable framework for rights limitations (Page and Soss Reference Page and Soss2021). This can be achieved by categorizing personal information into various levels and imposing distinct requirements on investigative agencies corresponding to each level of information. The safeguarding of human rights is paramount in this context. The classification of personal information takes into account several factors: (1) the necessity for individuals to share personal data with third parties, such as bank account details, to engage meaningfully in societal activities; (2) the presence of unique personal characteristics in the information, exemplified by medical records; (3) the ease with which third parties can access information that carries minimal expectations of privacy, such as content shared on social media; and (4) the existence of legal protections for certain types of information, like health-related data. Utilizing these four criteria, personal information can be categorized into varying levels of privacy, with corresponding protective measures established for each level. It is important to recognize that the advent of big data has significantly intensified the potential for privacy violations. The landscape of privacy in the era of big data differs markedly from that of previous times, necessitating the development of a new framework for the hierarchical protection of personal information that builds upon traditional standards of information protection.

Furthermore, regarding security concerns in the data storage process, it is evident from the practices of various countries that an organization may face administrative and/or criminal penalties if it fails to safeguard data security in accordance with established standards. In light of this, conducting a risk assessment of the big data storage process is a prudent measure for future preparedness. The key factors to consider include: (1) identifying potential threats that could affect the organization or arise from operations involving third-party agencies; (2) assessing the organization’s internal personnel, technology and resources to identify possible shortcomings; (3) evaluating the consequences of vulnerability-based attacks on individuals or organizations; and (4) estimating the probability of risks. Consequently, a dual regulatory approach focusing on both processes and outcomes can enhance the protection of personal information.