COLLEXA: An AI-Powered College ERP System with Embedded Academic Risk Prediction

Abstract

Everyworkingday,atypicalhigher-educationcam-pus produces thousands of structured records — session-level attendance, assessment grades, fee settlements, and course regis-trations—yetthesoftwareplatformsthatcollectthisdatararely exploit it to anticipate academic difficulty. This paper reportsthe design, construction, and empirical evaluation of COLLEXA (College Excellence), a web-native, intelligence-augmented En-terprise Resource Planning (ERP) framework that remedies this shortfall by weaving a machine-learning inference engine intothe heart of routine academic administration. The platform’s three-tier implementation pairs a React 18 / Tailwind CSS client with a FastAPI 0.110 application server and a MySQL 8.0 persistent store, with all communication secured through JSON Web Token (JWT) authentication and fine-grained Role-Based AccessControl(RBAC).AnembeddedAcademicRiskPrediction componentbuildsaper-studentfeaturetriplet—attendancerate, meanassessment score, andcumulativeGradePointAverage and submits it to a trained Random Forest ensemble that assignseachlearnertooneofthreeriskcategories:Low,Medium, or High. Tested against a stratified hold-out partition of 240 records, the ensemble reached 88.3% accuracy and a macro-averaged F1-score of 0.88; five-fold cross-validation over the full corpusreturned a mean of 88.3% ±0.6%, attesting to consistent generalisation. Under a simulated workload of 200 simultaneous users, the prediction endpoint delivered a 95th-percentile latency of 340 ms — well inside the 2,000 ms operational target — with zero request failures. All six stated development objectives were satisfied, and the system cleared every security, integration, and user-acceptance verification scenario. COLLEXA demonstrates that embedding predictive intelligence into day-to-day ERP operations transforms the system from an after-the-fact ledger into a real-time support mechanism for at-risk students.

Introduction

COLLEXA is an AI-integrated College ERP system designed to manage academic and administrative activities while predicting student academic risk in real time. Traditional college ERP systems handle tasks such as attendance, examinations, fee management, and course administration, but they lack predictive analytics to identify struggling students early. COLLEXA addresses this limitation by combining ERP functionality with a machine learning-based Academic Risk Prediction module.

The system provides a unified multi-role platform for students, faculty, administrators, and finance departments through a web-based application. It includes attendance tracking, assessment management, fee monitoring, dashboards, and RESTful APIs with secure JWT-based authentication and role-based access control. A QR-code attendance mechanism ensures secure and time-bound attendance recording.

The AI prediction component uses a Random Forest classifier trained on attendance percentage, average marks, and GPA to categorize students into Low, Medium, or High academic risk levels. Real-time prediction results are integrated directly into the ERP dashboards, enabling early intervention without requiring separate analytics tools. Attendance was identified as the most influential factor in predicting student performance.

COLLEXA follows a modular three-tier architecture consisting of:

• Presentation Tier: React and Tailwind CSS frontend
• Logic Tier: FastAPI backend with AI inference services
• Data Tier: MySQL database managed through SQLAlchemy ORM

The system emphasizes scalability, low latency, and security using HTTPS, bcrypt password hashing, JWT authentication, and strict role-based permissions.

Experimental evaluation showed strong performance:

• Random Forest achieved 88.3% accuracy and 0.88 macro F1-score
• It outperformed Logistic Regression and Decision Tree models
• Cross-validation confirmed stable and reliable predictions
• API response times remained below 340 ms even under 200 concurrent users
• All integration, security, and user acceptance tests passed successfully

Conclusion

This paper introduced COLLEXA, an intelligence-augmented College ERP framework built to resolve a pervasive limitation of conventional college information systems: their inability to convert the operational data they hold into timely, actionable indicators of student academic risk. Embedding a Random Forest classifier — trained on three live ERP features and queried through the same API layer that serves attendance records and grade reports — upgrades the platform from a passive data ledger into a proactive early-intervention instrument without imposing any new workflow burden on its users. Empiricalvalidationreturned88.3%classificationaccuracy, a macro F1-score of 0.88, and cross-validation stability of 88.3% 0.6%. Prediction latency under 200 concurrent users peaked at 340 ms, comfortably within the operational budget, and the system cleared every functional, security, and acceptance test. The entire stack runs on standard consumer-grade hardware, requiring no GPU or cloud infrastructure. Plannedfuturedirectionsinclude: 1) Mobile Client — A React Native or Flutter companion app will extend all role-specific views to smartphones, allowing faculty to generate QR codes and students to monitor their risk status from any location. 2) Sequential Deep Learning — LSTM and Transformer architectures will be explored to model the temporal evolution of student behaviour more precisely than a feature-aggregation approach. 3) Cloud Containerisation— Docker images and Kuber-netes orchestration will package the application for scal-able, fault-tolerant cloud deployment serving institutions with much larger student bodies. 4) Automated Alerts — Push notifications via SMS and emailwillbedispatchedautomaticallywheneverastu- dent’s predicted risk level escalates, enabling faculty to intervene without manually checking the dashboard. 5) Conversational Assistant — An LLM-backed chatbot embedded in the student dashboard will translate risk profiles into personalised, actionable study recommenda-tions. 6) Richer Feature Representation — Library usage records, counselling appointment logs, and extracurric-ular participation data will be incorporated to improve prediction granularity beyond assessment scores alone. These additions will mature COLLEXA into a compre-hensive, adaptive academic intelligence platform suitable for institutionsacrossabroadrangeofsizesandresourcelevels.

References

[1] T. H. Davenport, “Putting the enterprise into the enter-prise system,” Harvard Business Review, vol. 76, no. 4, pp.121–131,Jul.–Aug.1998. [2] R.PollockandJ.Cornford,“ERPsystemsandtheuniver-sity as a ‘unique’ organisation,” Information Technology & People, vol. 17, no. 1, pp. 31–52, 2004. [3] M. Nazim and B. Mukherjee, “Implementing library management systems in academic libraries: An analysis ofissuesandchallenges,”LibraryReview,vol.60,no.3,pp.256–268,2011. [4] C. Romero and S. Ventura, “Educational data mining: A review of the state of the art,” IEEE Trans. Syst., Man, Cybern. C, vol. 40, no. 6, pp. 601–618, Nov. 2010. [5] S. K. Yadav, B. Bharadwaj, and S. Pal, “Mining edu-cation data to predict student’s retention: A comparative study,”Int.J.Comput.Sci.Inf.Technol.,vol.2,no.2,pp.752–759,2011. [6] E.A.Amrieh,T.Hamtini,andI.Aljarah,“Preprocessing and analyzing educational data set using X-API for improvingstudent’sperformance,”inProc.IEEEJordan Conf. Appl. Electr. Eng. Comput. Technol. (AEECT), Amman, Jordan, 2015, pp. 1–5. [7] A.Daud,N.R.Aljohani,R.A.Abbasi,M.D.Lytras, [8] F. Abbas, and J. S. Alowibdi, “Predicting student per-formance using advanced learning analytics,” in Proc. 26thInt.Conf.WorldWideWebCompanion(WWW’17), Perth, Australia, 2017, pp. 415–421. [9] L.Breiman,“Randomforests,”MachineLearning, vol. 45, no. 1, pp. 5–32, Oct. 2001. [10] S. Ramirez, FastAPI: Modern, Fast Web Framework for BuildingAPIswithPython,Tiangolo,2023.[Online]. Available:https://fastapi.tiangolo.com [11] M. B. Jones, J. Bradley, and N. Sakimura, “JSON Web Token(JWT),”IETF,RFC7519,May2015.[Online]. Available:https://tools.ietf.org/html/rfc7519 [12] M.Bayer,SQLAlchemy—ThePythonSQLToolkit and Object Relational Mapper, SQLAlchemy.org, 2023. [Online]. Available: https://www.sqlalchemy.org [13] A. Wathan, Tailwind CSS — A Utility-First CSS Frame-work,TailwindLabs,2023.[Online].Available:https://tailwindcss.com [14] C.Piech,J.Bassen,J.Huang,S.Ganguli,M.Sahami,L. J. Guibas, and J. Sohl-Dickstein, “Deep knowledge tracing,” in Adv. Neural Inf. Process. Syst. (NeurIPS), vol. 28, 2015. [15] S. Pandey and G. Karypis, “A self-attentive model for knowledgetracing,”inProc.20thInt.Conf.Artif.In-tell.Educ.(AIED),Chicago,IL,USA,Springer,2019,pp.405–415.

Copyright

Copyright © 2026 M. Asha, B. Estheru Rani, G.S.S.M.V. Lakshmi Durga, K. Kavya. This is an open access article distributed under the Creative Commons Attribution License, which permits unrestricted use, distribution, and reproduction in any medium, provided the original work is properly cited.