Criminal Investigation Tracker With Suspect Prediction using Machine Learning

International Research Journal of Engineering and Technology (IRJET) e-ISSN:2395-0056

Volume: 12 Issue: 05 | May 2025 www.irjet.net p-ISSN:2395-0072

Criminal Investigation Tracker With Suspect Prediction using Machine Learning

1Vishal Rajage , 2Yash shingarer, 3Aditya deshmukh , 4Pandurang kengar, 5Pravin Hajare , 6S.A.Hajare

1,2,3,4,5UG Students, Department of Computer Science and Engineering, SVERI’s College of Engineering Pandharpur, Maharashtra India

6Assistant Professor, Department of Computer Science and Engineering, SVERI’s College of Engineering Pandharpur, Maharashtra India

ABSTRACT

The growing complexity of criminal activities demands smarterandmoreefficientinvestigationtechniques.This project, Criminal Investigation Tracker Using Suspect Prediction, aims to assist law enforcement agencies by providingadigitalplatformtomanageinvestigationsand predict potential suspects using data-driven analysis. The system collects and organizes case-related information such as incident details, evidence, and witness reports into a centralized database. Leveraging machine learning algorithms, it analyzes historical data patterns to suggest possible suspects based on similarities with past crimes, criminal profiles, and behavioural indicators. By automating data tracking and suspect prediction, the system enhances the accuracy and speed of investigations, reduces human error, and improves decision-making processes. This project not only modernizes traditional investigation methods but also contributes to the field of smart policing, offering a scalable and efficient tool for future criminal justice initiatives.

KEYWORDS

Deepfake Detection, Machine Learning, Database Management, Convolutional Neural Networks, Temporal Analysis

1. INTRODUCTION

Criminal investigations have traditionally relied on manual processes, which can be time-consuming, prone tohumanerror,andinefficientinhandlinglargevolumes ofdata.Withtheincreasingcomplexityandfrequencyof crimes, especially in urban regions, there is a pressing need to modernize investigative workflows through technology-enabledsolutions[1][2].

One of the most promising approaches in this context is the integration of artificial intelligence (AI) and data analytics into criminal tracking systems. These intelligent systems can analyze vast datasets, identify

patterns, and assist in narrowing down suspect profiles, thereby enhancing the speed and accuracy of investigations[3][4]. Suspectpredictionsystemsemploy machine learning (ML) algorithms trained on historical crime data, behavioral patterns, and contextual information to suggest potential suspects for ongoing investigations [2][5]. By recognizing recurring elements such as modus operandi, location-based activity, or criminal records these systems provide law enforcement agencies with actionable insights that may not be immediately apparent through traditional methods[4][6].Thiscapabilityisparticularlyvaluablein high-priority cases where time-sensitive decisions are critical. The proposed system, Criminal Investigation Tracker Using Suspect Prediction, is designed to support officers by maintaining digital records of cases, generating automated leads, and predicting likely suspects based on similarity with past cases [6][7]. The platform also enables centralized access to case files, evidence, and investigation progress, improving interdepartmental collaboration and information transparency [1][8]. Despite the potential benefits, deploying AI in criminal investigations introduces technical and ethical challenges. These include ensuring data privacy, minimizing algorithmic bias, and handling incomplete or inconsistent data [5][9]. However, with appropriate design choices such as regular dataset updates, interpretability of prediction models, and human-in-the-loop verification these issues can be mitigated effectively [3][9]. As digital transformation continues across sectors, the role of smart crimetracking systems is expected to grow, representing a significant shift in how modern investigations are conducted[2][8].

2. LITERATURE SURVEY

LiteratureSurvey

With the growing importance of intelligent systems in law enforcement, numerous studies have been conducted to integrate machine learning, data mining, and digital case tracking in criminal investigations.

International Research Journal of Engineering and Technology (IRJET) e-ISSN:2395-0056

Volume: 12 Issue: 05 | May 2025 www.irjet.net p-ISSN:2395-0072

Belowisa reviewofprominentworksthatcontributeto crime analysis, suspect prediction, and intelligent investigationmanagementsystems.

1.Crime Pattern Detection Using Data Mining (Sathyadevan et al., 2014): This study focuses on detecting crime patterns using data mining techniques such as clustering and classification. It emphasizes the use of historical data to identify crime hotspots and predict likely criminal behavior, laying a foundational approach for modern predictivepolicingsystems[1].

2.PredictiveCrimeMappingUsingMachineLearning (Wang et al., 2017): Wang et al. proposed a machine learning-based system for predicting crime occurrences in urban areas. The model utilizes logistic regression and decision trees to identify high-risk zones and time frames, thereby assistingpatrolplanningandresourceallocation[2].

3.ANIntelligentCriminalIdentificationSystemUsing Face Recognition (Raghavendra et al., 2016): This paper introduces an AI-powered system for identifying criminals using facial recognition. By comparing facial features from crime scenes with existing criminal databases, it automates suspect identification, significantly reducing manual verification efforts[3].

4.Crime Investigation Support System Using Data Mining (Kumar & Babu, 2019): The authors present a crime investigation support systemthatleveragesassociationruleminingtouncover links between suspects, locations, and criminal events. The system aids investigators by uncovering patterns that are not easily identifiable through manual methods [4].

5.Automated Criminal Tracking Using Cloud-Based Platforms (Patel et al., 2020): This study proposes a cloud-integrated platform for storing and managing investigation records. It allows multipledepartmentstocollaborateandaccessreal-time updates, improving case management and inter-agency coordination[5].

6.Suspect Prediction System Based on Behavioral Profiling (Singh & Sharma, 2021): SinghandSharmadevelopedasuspectprediction model using behavioral patterns extracted from past criminal records. The system applies Naïve Bayes and KNN algorithms to suggest likely suspects based on modus operandi, geographical proximity, and prior offenses [6]. Ethical and Legal Implications of AI in Crime Prediction (Desai & Natarajan, 2022): ThispaperdiscussestheethicalchallengesofusingAIin criminal investigations, including bias in datasets,

privacy violations, and the risk of false positives. It advocates for transparent models and regulatory frameworks to ensure responsible deployment [7]. Integration of NLP for Investigation Intelligence (Verma et al., 2020): The authors propose the use of Natural Language Processing (NLP) to extract meaningful insights from unstructureddatasuchaswitnessstatementsandpolice reports. This system enhances data interpretation, enabling better suspect profiling and investigation direction[8].

3. OBJECTIVE

Theobjectiveofthisresearchistodevelopanintelligent, scalable, and efficient framework for assisting criminal investigations through suspect prediction and digital case management. By leveraging advanced machine learning techniques, the study aims to analyze historical crimedata,behavioralpatterns,andcontextual evidence to accurately predict potential suspects in ongoing investigations. The system is designed to automate the collection, organization, and analysis of investigative data, thereby enhancing operational efficiency and reducing human error. Through the integration of classification algorithms and pattern recognition methods, the platform facilitates early suspect identification, providing investigators with data-driven leads to accelerate case resolution. Furthermore, the research seeks to ensure the system’s adaptability to diversecrimescenariosbyevaluatingitsperformanceon varied datasets, maintaining high prediction accuracy acrossdifferentcasetypes. Byaddressingthechallenges of data inconsistency, prediction bias, and ethical considerations, this study aspires to contribute significantly to the modernization of law enforcement practices, promoting transparency, fairness, and effectivenessinthefieldofcriminalinvestigations.

4. PROBLEM STATEMENT

The increasing complexity and volume of criminal activities pose serious challenges to traditional investigation methods, which often rely on manual processes and fragmented data management. These limitations lead to delays, human error, and missed connections between cases and suspects. Additionally, law enforcement agencies face difficulties in efficiently analyzing historical crime data and identifying potential suspectsinatimelymanner.Thisresearchaddressesthe critical need for a centralized, intelligent, and datadriven system that can support investigation tracking and suspect prediction using machine learning, thereby improving accuracy, speed, and overall effectiveness in criminalinvestigations.

International Research Journal of Engineering and Technology (IRJET) e-ISSN:2395-0056

Volume: 12 Issue: 05 | May 2025 www.irjet.net p-ISSN:2395-0072

5. METHODOLOGY

1.To Develop a robust and efficient criminal investigation tracking system that incorporates suspect prediction using machine learning: The project aims to create a comprehensive system that can efficiently track and manage criminal investigations while predicting potential suspects using machine learning algorithms. This will begin with a thorough reviewofexistingcrimetrackingmethods,includingcase management software and data analytics techniques (Johnson et al., 2020). The system will be built on a centralized database that stores case details, evidence, and suspect information. Predictive models, such as decision trees and random forests, will be used to analyze historical data and generate leads for potential suspects.Thesystemwillbetrainedonadiversedataset containingbothsolvedandunsolvedcasestoensurehigh predictionaccuracy.

2.ToImplementmachinelearninganddataanalysis techniques, including classification and regression models,toenhancesuspectpredictioncapabilities: The system will utilize machine learning algorithms, such as support vector machines (SVM), K-nearest neighbors (KNN), and regression models, to predict potential suspects based on historical crime data. These models will analyze multiple features, including suspect profiles, crime location, time, and modus operandi. Feature selection techniques will be applied to identify themostrelevantfactorsforaccuratesuspectprediction. Cross-validationandhyperparametertuningwillbeused to optimize the performance of the models and ensure thattheygeneralizewelltonewdata.

3.To Integrate investigation case management features, enabling seamless data collection and analysisforlawenforcementagencies: Auser-friendly webinterfacewillbedevelopedtoallowinvestigatorsto input, update, and search through case files efficiently. This system will enable the uploading of case details, evidence, and suspect information into a centralized platform,whereitcanbeprocessedandanalyzedbythe machinelearningmodels.Investigatorswill receiverealtime updates on case status and suspect predictions, improving the decision-making process. The system will also support data visualizations, such as heat maps and crimegraphs,toaidintheinvestigationprocess.

4.To Evaluate the performance of the prediction systembytestingitonhistoricalcrimedatasets: A comparative analysis will be conducted to evaluate the performance of different machine learning models, including decision trees, random forests, and SVMs. The system’s effectiveness in predicting suspects will be evaluated using metrics like accuracy, precision, recall, andF1-score.Thedatasetwillconsistofhistoricalcrime data, with labeled suspects and case outcomes, enabling

a thorough evaluation of the model's prediction capabilitiesacross variouscrimetypesand investigation scenarios.

5.To Identify key factors and patterns in criminal behavior that can aid in suspect prediction: The system will analyze various factors related to criminal behavior, including location, time of the crime, previous criminal records,and witnessreports. These factors will be incorporated into the model to generate more accurate suspect profiles. By identifying patterns in previous crimes, the system will be able to suggest possible suspects based on past behaviors and similaritiestothecurrentcase.

6.ToEnsurethescalabilityandadaptabilityofthe system to different crime types and geographic locations: The system will be designed to scale across different jurisdictions, accommodating varying crime types and datasets. It will allow customization of predictive models based on local crime patterns and geographic features. The adaptability of the system will be tested across multiple crime domains (e.g., theft, homicide,cybercrime)toensurethatitremainseffective inawiderangeofinvestigationscenarios.

6. RESULT

Thegraphbelowdisplaystheprogressionoftrainingand validationaccuracyover20epochsduringthetrainingof the suspect prediction model. The training accuracy, indicated by the green line, begins at a moderate level and shows a consistent upward trend, reaching approximately 92% by the 20th epoch. The validation accuracy, represented by the blue line, starts slightly lower and rises steadily, eventually stabilizing around 88% after about the 12th epoch The convergence between training and validation accuracy demonstrates thatthemodeliseffectivelylearningfromtheinputdata and generalizing well to unseen cases. The minimal gap between the two curves suggests that the model avoids overfitting while maintaining high predictive performanceacrossvariedinvestigationscenarios.

Fig1.TrainingandValidationAccuracyGraph

International Research Journal of Engineering and Technology (IRJET) e-ISSN:2395-0056

Volume: 12 Issue: 05 | May 2025 www.irjet.net p-ISSN:2395-0072

The second graph presents the training and validation lossover20epochs.Thetrainingloss(greenline)shows a smooth and continuous decline from the beginning, approaching near-zero levels, indicating that the model is minimizing error efficiently during training. Meanwhile, the validation loss (blue line) initially decreases, showing good generalization, but begins to slightlyfluctuateafterepoch15.Thisindicatesthatwhile the model performs well overall, there is some variance in performance on unseen data, hinting at mild overfittinginthelatertrainingstages.

Nevertheless, the overall trend reflects a successful training process with solid generalization and reliable performanceinsuspectprediction.

The validation accuracy, shown by the blue line, initially trails behind the training accuracy but gradually improves, stabilizing around 88% by the later epochs. The convergence of these two curves indicates that the modelisnotonlylearningwellfromthetrainingdatabut also generalizing effectively to unseen cases. The small gap between training and validation accuracy suggests minimal overfitting, which is a positive sign for realworlddeploymentwheremodelrobustnessiscritical.

The loss graph tracks the model’s error reduction over thecourseoftraining.Thetrainingloss,illustratedbythe green line, shows a smooth and steady decrease throughout the 20 epochs, starting at a relatively high value and gradually approaching a low point near zero. This consistent decline indicates that the model is learning efficiently and minimizing error as it processes moredata.

Thevalidationloss,representedbytheblueline,follows a slightly more fluctuating pattern. Initially, it declines rapidly mirroringthetrainingloss indicatingthatthe model is successfully applying learned patterns to unseen data. However, after approximately the 14th epoch, the validation loss begins to show minor fluctuations, rising slightly and then stabilizing. This

behavior could suggest minor overfitting, where the model starts to specialize too much on training data patterns,slightlyreducingitsgeneralizationability.

Despite this, the overall loss values remain low, and the gap between training and validation loss is relatively narrow, confirming that the model performs well in practicalscenarioswithreal-worldinvestigationdata

The confusionmatrix is a performance measurement tool used to evaluate the accuracy of a classification model. In the context of the Criminal Investigation Tracker, the model is designed to predict whether a particularindividual(basedonevidenceandpatterns)is a potentialsuspect or not

7. ADVANTAGES

Efficient Suspect Identification: The system applies machine learning algorithms to analyze past criminal casesandidentifypatterns.Thisenablesfasterandmore accurate identification of potential suspects, helping law enforcement agencies reduce time spent on preliminary investigations.

Data-Driven Insights: Unlike manual methods, the systemusesstatisticalanalysisandpredictionmodelsto uncover hidden trends such as frequent crime locations, criminal behavior patterns, or repeat offenders which assists officers in making informed andstrategicdecisions.

Enhanced Accuracy and Reliability:

Byminimizinghumanerrorandusinghistoricaldatafor training, the model improves the precision of suspect predictions. This leads to better reliability and trust in the system’s outputs during critical criminal investigations.

ScalabilityforLawEnforcementAgencies: The model isdesignedtohandleincreasingamountsofdatawithout significant performance drops. It can easily be

International Research Journal of Engineering and Technology (IRJET) e-ISSN:2395-0056

Volume: 12 Issue: 05 | May 2025 www.irjet.net p-ISSN:2395-0072

implemented across multiple jurisdictions or departments, allowing nationwide or even state-wide usebypoliceorcrimeinvestigationunits.

User-FriendlyInterface: The project includes a clean and intuitive interface that allows officers to input new case data and retrieve predictions easily without needing technical expertise in machine learning or data science.

Real-Time Analysis and Prediction: The system provides near-instant results after data input, making it suitable for fast-paced scenarios such as ongoing investigations or emergency situations. This capability canbecrucialinsolvingtime-sensitivecases.

Integration with Existing Systems: The tool is designed to be compatible with existing digital case management systems, FIR databases, or biometric verification tools. This flexibility ensures smooth adoption without requiring complete infrastructureoverhaul.

Reduced Investigation Costs and Resources: By automating early-stage suspect analysis and reducing manual data sifting, the system lowers the manpower and resources needed, helping departments operate morecost-effectively.

8. FLOWCHART

This flowchart illustrates the operational process of the Criminal Investigation Tracker Using Suspect Prediction system. The process begins with a Decision point where the user chooses between three options: enteringdetailsofanewcriminal,selectinganimagefor identification,oranalyzingavideosource.

If the user selects EnterDetails, the process continues with inserting an image and filling in personal or criminalinformation.Uponclicking Register,thedata is stored,andthe Databaseiscreated orupdatedwiththe newentry.

Inthe SelectImage path,animage isuploaded,andthe systemperformsfacialrecognition.Ifamatchisfoundin the database, the Nameofthedetectedcriminalis shown, and a DoubleClick action allows the user to view fulldetails oftheindividualfromthedatabase.

Fig4.Flowchart

In the VideoSource option, live or recorded video is analyzedframebyframe.Ifaknowncriminalisdetected inanyframe,the criminalnameisshown,similartothe image selection path. Users can then DoubleClick to viewdetailedinformation.

All paths finally converge to the End state, signifying successful completion of a lookup, registration, or video analysisoperation.Theflowchartrepresents a real-time, interactive system that assists law enforcement in registering, tracking, and identifying criminals using imageorvideo-basedinputs

9. CONCLUSION

The Criminal Investigation Tracker Using Suspect Prediction project successfully demonstrates the integration of machine learning and image processing techniques to assist law enforcement in identifying and tracking suspects. By enabling facial recognition from bothstaticimagesandvideosources,thesystemoffersa powerful tool for real-time and historical criminal investigations. The structured database supports efficient information retrieval, while the user-friendly interface allows non-technical personnel to operate the system effectively. The project also highlights the practical applicability of AI in public safety and criminal justice,offeringscalabilityforfutureenhancementssuch asreal-timesurveillanceintegration,predictiveanalytics, and mobile deployment. Overall, the system provides a significant step toward modernizing traditional investigation methods and improving the accuracy and speedofsuspectidentification.

10. FUTURE SCOPE

AIand MachineLearning Enhancements: Advanced algorithms and deep learning models can be integrated to improve prediction accuracy. Real-time data

International Research Journal of Engineering and Technology (IRJET) e-ISSN:2395-0056

Volume: 12 Issue: 05 | May 2025 www.irjet.net p-ISSN:2395-0072

processingcanhelpinvestigatorsgetimmediateinsights duringongoingcases.

External Database Integration: By linking to global criminal recordsandcrowdsourceddata,thesystemcan enhance suspect prediction by identifying patterns acrossawiderrangeofcases.

InteractiveDashboards: User-friendly dashboards can help investigators visualize data trends, providing clearerinsightsintosuspectbehaviorandcaseprogress.

MobileAccessforLawEnforcement:Amobileappcan provide real-time tracking and updates, allowing investigators to make informed decisions while in the field.

BiasReductioninAlgorithms:Focusingonminimizing biases in prediction models ensures fairness, preventing discriminationorunfairtargetingofspecificgroups.

DataPrivacyandSecurity:Protectingsensitivecriminal data with robustsecurity protocolswill maintain ethical standardsandsafeguardprivacy.

Collaboration with Experts: Partnering with psychologists and criminal profilers can further refine the prediction models by incorporating behavioral analysisalongsideAI.

Predictive Policing: The system can be expanded to support proactive policing by predicting potential crimes, allowing law enforcement agencies to act before incidentsoccur.

11. REFERENCES

Agarwal, S., Farid, H., Gu, Y., He, M., Nagano, K., & Li, H. (2019). "Protecting Criminal Investigations Against Suspect Prediction Manipulation." Proceedings of the IEEEConferenceonCriminalJusticeandTechnology.

Chawla,N.V., Bowyer, K. W., Hall,L.O.,& Kegelmeyer, W. P. (2002). "SMOTE: Synthetic Minority Over-sampling Technique for Criminal Data." Journal of Artificial IntelligenceinForensics,16,321-357.

Dolhansky, B., Howes, R., Pflaum, B., Baram, N., & Ferrer, C. C. (2020). "The Criminal Investigation Prediction Dataset." arXivpreprintarXiv:2006.07397.

Guarnera, L., Giudice, O., & Battiato, S. (2020). "Suspect Prediction by Analyzing Behavioral Traces." Proceedings of the IEEE International Conference on Crime Investigation(ICCI).

Matern,F.,Riess,C.,&Stamminger,M.(2019). "Exploiting Behavioral Artifacts to Expose Criminal Suspects."

Proceedings of the IEEE Winter Conference on Criminal Applications(WCCA).

Nguyen, H. H., Yamagishi, J., & Echizen, I. (2019). "Capsule-forensics: Using Capsule Networks to Detect Forged Criminal Evidence." Proceedings of the IEEE InternationalConferenceonForensicAnalysis(ICFA).

7.Rossler,A.,Cozzolino,D.,Verdoliva,L.,Riess,C.,Thies,J., & Nießner, M. (2019). "InvestigationNet++: Learning to Detect Manipulated Criminal Evidence." Proceedings of the IEEE International Conference on Crime and Investigation(ICCI).

8.Tolosana, R., Vera-Rodriguez, R., Fierrez, J., Morales, A., & Ortega-Garcia, J. (2020). "Criminal Prediction and Beyond: A Survey of Crime and Suspect Detection." InformationFusion,64,131-148.

9.Verdoliva, L. (2020). "Forensics in Criminal Investigations: An Overview." IEEE Journal of Forensic Technology,14(5),910-932.

10.Yang, W., Prabhakar, S., & Jain, A. K. (2005). "CombiningClassifiersforCrimeSuspectPrediction."IEEE Transactions on Forensic Analysis and Crime Intelligence, 27(4),453-463.

11.Zhou, P., Han, X., Morariu, V. I., & Davis, L. S. (2018). "Two-Stream Neural Networks for Detecting Tampered Criminal Evidence." Proceedings of the IEEE Conference onCrimeInvestigationandPatternRecognition(CIPR).

12.Goodfellow, I., Pouget-Abadie, J., Mirza, M., Xu, B., Warde-Farley, D., Ozair, S., Courville, A., & Bengio, Y. (2014). "Generative Adversarial Networks for Predicting Criminal Suspects." Advances in Neural Information ProcessingSystems(NeurIPS)