DOI : 10.17577/IJERTCONV14IS110145- Open Access

- Authors : Deepika S, Aparna D, Bharathi Gr
- Paper ID : IJERTCONV14IS110145
- Volume & Issue : Volume 14, Issue 11, CONFLUENCE – 2026
- Published (First Online) : 20-11-2026
- ISSN (Online) : 2278-0181
- Publisher Name : IJERT
- License:
This work is licensed under a Creative Commons Attribution 4.0 International License
Foodhub, Ai Based Smart Surplus Food Redistribution System
Deepika S1, Aparna D2, Bharathi GR3
Student, Department of BCA, St Francis College, Bengaluru1 Student, Department of BCA, St Francis College, Bengaluru2
Assistant Professor, Department of BCA, St Francis College, Bengaluru3
Email : bharathi.gr7@gmail.com, deepikas5106@gmail.com, aparna0630@gmail.com
Abstract – Food waste is commonly seen in hotels, restaurants, hostels, bakeries, and during functions. During and after events, a considerable amount of food remains, often discarded despite its potential for consumption. At the same time, many NGOs and other organizations need food for people who require it. Often, the main difficulty is finding a quick and convenient way to connect donors with these organizations.
In this paper, we present AI-Based Smart Surplus Food Redistribution System, as a website to simplify this process. A donor can upload information about the available food, including its type, quantity, preparation time, expiry time, and pickup location. The system uses those details to assign a priority to the donation and helps in finding a nearby NGO for collection.
The system also allows users to check the progress of a donation after it has been accepted. Pickup and delivery details can be updated, while QR code verification can be used during the food handover. Donors can also earn reward points when they make successful donations, which encourages them to donate surplus food again. An admin panel is available to manage users, donations and other activities of the website. The system is also designed so that users can access it easily from mobile devices.
The main aim of the proposed system is to reduce the unnecessary disposal of usable food and make the process of connecting food donors with NGOs simpler, faster, and more organized.
Keywords: Artificial Intelligence, Food Waste Reduction, Surplus Food Redistribution, NGO Matching, Sustainable Food Management
-
INTRODUCTION
Food waste is pervasive across the food service and hospitality sector, where hotels, restaurants, hostels and bakeries over-prepare to meet uncertain demand, and where functions, banquets and other organized events frequently leave considerable quantities of edible food. Globally, an estimated 1.05 billion tons of food were wasted in 2022 at the retail, food service and household levels- roughly 19 per cent of all food made available to customers- with food service accounting for 28 percent and retail for 12 per cent of this waste, while a further 13.2 per cent of food is lost in the supply chain between harvest and retail (UNEP, Food Waste Index Report 2024; FAQ, Food Loss & Food Waste).
The central problem is therefore not scarcity but coordination: much of this surplus remains safe and edible for a short window, yet it is discarded because donors have no fast, reliable channel to reach organizations that could use it. In parallel, an estimated
8.2 per cent of the worlds population- approximately 673 million people- experienced hunger in 2024 (FAQ). Existing recovery efforts depend largely on informal telephone calls, personal contacts and messaging groups, which are slow, inconsistent and offer no record of quantity, preparation time, expiry time or handover. As a result, donations are missed, collected too late or accepted without sufficient information for the receiving NGO to plan distribution.
This paper presents the AI-Based Smart Surplus Food Redistribution System, a mobile-accessible website that connects food donors with nearby NGOs. A donor uploads the food type, quantity, preparation time, expiry time and pickup location; the system used these attributes to assign a priority to the donation and identify a suitable nearby NGO for collection. Once accepted, the
donation can be tracked, pickup and delivery details updated, and QR code verification performed at handover, while donors earn reward points for successful donations and an admin panel manages users and platform activity. The system aims to reduce the unnecessary disposal of usable food and make redistribution simpler, faster and more organised.
Key contributions of this paper are:
-
A formulation of surplus food redistribution as a time-sensitive coordination and traceability problem rather than a simple listing task.
-
A priority assignment mechanism that derives urgency from donor-supplied attributes including food type, quantity, preparation time and expiry time.
-
A nearby-NGO matching approach that uses pickup location to route each donation to an appropriate receiving organisation.
-
An end-to-end donation lifecycle combining status tracking, handover verification via QR code, donor reward points and an administrative panel.
-
-
LITERATURE REVIEW
SLNO
Author(s)
Year
Title
Method/ Approach
Key Findings
Limitations/ Gaps
1
Vlachogianni, A.
2026
Food Redistribution:A Case Study on the Bring It Back Platform[1]
Case study of a digital redistribution platform
Digital platforms are reshaping how surplus perishable food is
redistributed
Single case; no ML model ore metrics
2
Dionisi, L
2026
FoodLink AI: Artificial Intelligence and Digital Platforms for reducing food waste through food redistribution[2]
AI+Digital- Platform concept for redistribution
Concludes that AI-driven platforms can reduce food waste by turning surplus food into organised
donations
Conceptual only; no model comparison, dataset or measured accuracy
3
Liu, M. ;
Costa, U.
2026
Artificial Intelligence applications for optimizing food distribution and reducing community food
waste[3]
Review of AI methodologies for food distribution
Converts ML for demand forecasting and predictive analytics
Preprint- No confirm venue or peer review
4
Madhu
2025
AI-Driven food packaging Systems[4]
AI-driven packaging / shelf-life intelligence
Describes AI systems predicting shelf life or contamination from historic data and stresses that temperature fluctuations in cold chains drastically affect
perishables
Packaging/Sen sor-centric: requires instrumentatio n unavailable to a plain donation web form
5
Seilov,Abildino v and Sutula
2025
Smart innovations in food spoilage detection[5]
ML+Iot sensing (electronic nose); reinforcement- learning agent
Uses food storage conditions and spoilage rates to predict likelihood of spoilage; discusses an RL agent rewarded for actions that preserves
shelflife
Depends on IOT hardware that donors do not have; not deployable through a lightweight web API
6
Lytras, Lyberatos and Lytras
2025
Development of a model composting process for food waste in an Island community and use of machine learning models to predict its performance[6]
Five ML models including linear regressions
Employs five distinct L models to predict values associated with the composting process and accesses
feasibility
Treats food waste after disposal (composting), not prevention by redistribution; no urgency classifications
7
Beshiri
2023
Authentication and authorization
management in
Access lists and role-based access
Examine authentication
and
Enterprise SOA
context;no
SOA with the focus on RESTful services[7]
control;JWT tokens
authorization for restful services,using access list or RBAC to decide access and JWT as a
token format
lightweight application incrementatio n or domain evaluation
8
Kornienko and Mishina
2021
Principles of securing RESTful API web services developed with python frameworks[10]
Flask RESTful API
prototype;JWT as access- control mechanism
Builds a RESTfulAPI in python with flask and adapts JSON with tokens as the main authentication or access- control
mechanism
Security- focus; does not address multi-role authorization for a domain workflow such as donation approval
9
Makin and Abdullah
2025
Reducing food waste in culinary institutions[8]
Linear Regression
Argues demand forecasting is vital for minimizing food waste and applies linear regression to inventory management
Linear model with limited capacity for non-linear interactions; institution- scope inventory rather than surplus
redistribution
10
Naeem, Isnam, Barua and Araf
2026
A machine learning framework for predicting restaurant food waste to support sustainable food management[9]
ML Framework; demand forecasting under variable conditions
Develops a framework for predicting restaurant food waste and reviews the value of demand
forecasting
Predict waste volumes for a business; does not classify donation urgency or serve a donor- facing
workflow
under variable
conditions
-
METHODOLOGY
-
System Overview
FOOD HUB is a Flask-based web application (SQL Alchemy over a relational database, HTML/CSS/JavaScript front end, REST APIs) organised as a sequential pipeline: authentication establishes identity and role, the donation module captures the food record, the freshness module derives a freshness score, the machine-learning module converts it into an urgency class, the matching module ranks recipient organisations, and the NGO, volunteer, and handover, reward and monitoring modules complete the cycle. Only priority prediction is learned from data; freshness, distance and matching use explicit formulas and weighted scoring rules.
-
User Registration and Role-Based Authentication
Four roles are supported: Donor, NGO, Volunteer and Admin. Users register with role-specific attributes and log in to an authenticated session. Role-Based Access Control is enforced at the API level: a donor sees only their own donations, an NGO only donations visible to it, a volunteer only assigned deliveries, and the administrator holds system-wide read access.
-
Food Donation and Data Acquisition
An authenticated donor submits food name/type, category, quantity, preparation and expiry times, pickup location with latitude and longitude, and an optional description. The record is persisted with a unique donation identifier and creation timestamp, which later stages use to compute donation age and remaining expiry time.
-
AI-Based Food Priority Prediction
Priority prediction is the only machine-learning component. A Random Forest classifier is trained on food category, quantity, donation age A, remaining time R and the freshness score F, and outputs one of four classes: Low, Medium, High or Critical. The ensemble of K trees, each trained on a bootstrap sample with random feature subsampling, assigns the majority class:
(x) = {, , , } ( = 1 ) 1[_() = ]
The class-probability estimate ( | ) = (1/) 1[_() = ] gives confidence in the prediction. The predicted class sets the handling order of donations and supplies the urgency term used by matching.
-
Smart NGO Matching
Matching is deliberately not a machine-learning model but a deterministic, score-based ranking mechanism. For a candidate NGO n, four normalised components are combined: distance D(n), higher for nearer organisations; requirement fit Q(n) between the food category and the NGOs declared dietary requirements; capacity C(n) for the offered quantity; and urgency U(n), mapped from the predicted priority (Low < Medium < High < Critical). The matching score is the weighted sum
S(n) = w_d D(n) + w_r Q(n) + w_c C(n) + w_u U(n), w_d + w_r + w_c + w_u = 1
Eligible organisations (available capacity, distance within the service radius) are ranked by descending S(n), and the highest-ranked are presented as suitable recipients. Distance used here and in delivery, is the great-circle Haversine distance computed from latitude and longitude (, ) over the Earths mean radius R. The learned urgency signal is thus one weighted term in the rule-based score; distance, requirement fit and capacity remain purely algorithmic.
-
NGO Request and Claim
NGOs view the surfaced donations and request or claim one. The system verifies that the donation is still available and that the NGO is authorised before confirming, after which the donation moves to a claimed state for volunteer assignment.
-
Volunteer Delivery and Route/Distance Optimization
A volunteer transports the claimed donation from the pickup location to the NGO. Distance is obtained from the Haversine formula and an estimated travel time is derived as t_travel = d / v_avg is a configurable average speed; no external mapping service or continuous positioning is used.
-
QR-Based Handover and Donation Completion
A QR code encoding the donation identifier is generated at the point of transfer and scanned to confirm the physical handover. Only after the scanned identifier is validated against the database record is the donation marked delivered and the transaction closed.
-
Reward System
On successful completion, reward points are credited to the donors account as a rule-based acknowledgement of the contribution, encouraging sustained participation rather than competitive ranking.
-
Administrative Monitoring
Through a centralized interface the administrator oversees users, donations, NGO activities, volunteers, delivery status and overall system activity, so pending claims, incomplete deliveries and anomalies can be identified and resolved.
-
-
RESULTS AND DISCUSSION
-
Model Training Outcome
The food-priority classifier was trained with backend/ml/train/model.py using a RandomForestClassifier over six features (expiry_hours, quantity, food_type, donation_age_hours, storage_condition, food category) with a ColumnTransformer+OneHotEncoder preprocessing stage. Training succeeded and produced three artefcts: the serialised model(food_priority_model.pkl), its metadata (model_metadata.json), and the dataset (training_data.csv).
Metric
Reported Value
Validation accuracy
0.8750 (87.5%)
Training status
Successful; model and metadata persisted
Features
6(3 numerical, 3 categorial)
Output classes
4- Low, Medium, High, Critical
-
Food-Priority Prediction API
Post / api / food-priority / predict returns a priority label, confidence and class probabilities (frontend
Flask validation ml / predict.py model). The documented example (expiry_hours = 2, quantity
= 10, cooked, age 1 h, refrigerated, non_vegetarian) returned Critical, confidence 0.91; P(Critical/High/Medium/Low) = 0.91 / 0.06 / 0.02 / 0.01. Confidence equals the predicted classs probability a vote share, not a calibrated score. Incomplete payloads are rejected server-side with the missing fields named, so validation guards the inference path independently of client-side checks.
-
End-to-End Functional Results
Using the seeded demo accounts, all four roles logged in (JWT via Flask-JWT-Extended) and reached only their own functions: donors created donations, which received a priority label with confidence; NGOs listed and claimed donations; volunteers managed pickups and delivery status; administrators accessed users, donations, deliveries and analytics confirming role-based access control is enforced in the running system, not merely declared.
-
Automated Testing and Deployment
python -m pytest -q reported 4 passed, 0 failed (tests/test_auth_roles.py, tests/test_food_priority.py; per-module counts and coverage: [insert value]); malformed inputs, expired tokens and concurrent claims are not covered. The Flask backend runs at http://127.0.0.1:5000 (/api/health reachable), the frontend at http://localhost:8000, with SQLite under backend/instance/ via Flask- SQLAlchemy; re-seeding by removing instance/*.db and running python seed.py. Latency, throughput and concurrency were not measured ([insert value] if benchmarked).
-
Discussion
-
87.5% validation accuracy comfortably exceeds the ~25% random baseline, and tree ensembles are well supported for tabular data of this kind (Grinsztajn et al.; Uddin and Lu; Ali et al.). The models value is the routing it enables: labels let NGOs triage and volunteers sequence pickups, while the probability vector makes decisions auditable. Caveats: 0.91 is a vote share, not a calibrated probability, and nominal treatment of the ordinal labels penalises a Critical-for-High error like a Low-for-High one. Comparable donor-NGO-volunteer platforms are evaluated mostly functionally; this projects contribution is a trained, API-served, test-validated classifier as a live decision path, though matching remains claim-based unlike Shi et al. 's recommender approaches.
CONCLUSION
FOOD HUB addresses the twin challenges of perishability and coordination through a sequential pipeline: role-based registration, donation capture, freshness analysis, machine-learning priority predication, score-based NGO matching, claim and volunteer delivery, QR-verified handover, reward crediting and administrative monitoring. The design is deliberately hybrid: the Random Forest classifier is the only learned component, converting food category, quantity, donation age, remaining time and freshness score into a four-level urgency class, while freshness quantification, Haversine distance computation and NGO ranking rely on explicit formulas and weighted, rule-based scoring. The predicted priority feeds matching as an urgency term, so the most perishable donations are handled first and matched to the organisation best placed to receive them. Every stage produces explicit, inspectable data and handover is QR-validated, making the loop traceable, transparent and verifiable, while rewards and monitoring sustain participation and oversight. Future work includes real-world deployment, learning the matching weights from historical data, traffic- aware routing, and image-based freshness assessment.
REFERENCE
-
A. Vlachogianni, Food Redistribution : A Case Study on the Bring It Back Platform, SSRN Working paper, art. no. 6954058,2026. doi:10.2139/ssrn.6954058
-
L. Dionisi, FoodLink AI: Artificial Intelligence and digital platforms for reducing food waste through food redistribution, SSRN Working paper, art. no 6893198, 2026. doi:10.2139/ssrn.6893198
-
M. Liu, U. Costa, and R. Jav, "Artificial Intelligence Applications for optimizing food distribution and reducing community food waste, Preprint, 2026.
-
B. Madhu, AI-driven food packaging systems: a new frontier in intelligent food safety and shelf-life management, Journal of Food Science, 2025. doi:10.1111/1750-3841.70716
-
S. Seilov, D. Abildinov, M. Y. Sutula, A, Smart innovations in food spoilage detections: A focus on electronic nose, machine learning and IoT for perishable foods, Journal of Robotics and Control (JRC), vol. 6, no. 3, art. No. 25792, 2025. doi:10.18196/jrc.v6i3.25792
-
C. Lytras, V. Lyberratos, G. Lytras, Development of a model composting process for food waste in hand island community and use of machine learning models to predict its performance, Waste and Biomass Valorization, 2025. doi:10.1007/s12649-024-02697-9
-
A. Beshiri, Authentication and authorization management in SOA with the focus on RESTful services, International Journal of Software Engineering and Knowledge Engineering, 2023. doi:10.1142/S0218194023500328
-
A. H. Makin and M. I. Abdullah, Reducing food waste in culinary institutions: a predictive model for inventory management using linear regression, in Proc. 2025 IEEE 15TH Conf., 2025. doi:10.1109/ISCAIE64985.2025.11080771
-
M. M. H. Naeem, M. A. Islam, M. Barua, and I. A. Araf, A machine learning framework for predicting restaurant food waste to support sustainable food management, arXiv.2609.08078[cs.LG], 2026. doi:10.48550/arXiv.2609.08078
-
D. V. Kornienko and S. V. Mishina, Principles of securing RESTful API web services developed with python frameworks, Journal of Physics: Conference Series, vol. 2094, no.3, art. No. 032016, 2021. Doi: 10.1088/1742-6596/2094/3/032016.
