Literature Review
Urban Waste Crisis
The World Bank projects that global solid waste generation will reach 3.4 billion tons annually by 2050, a 70% increase on 2016 levels. In Sri Lanka, the Colombo Municipal Council alone generates approximately 800 tons of municipal solid waste per day, and existing infrastructure operates well below the level needed for effective management.
IoT in Waste Management
Low-cost IoT platforms, particularly ESP32-class microcontrollers with integrated Wi-Fi, and cloud services such as Firebase make intelligent, data-driven waste management feasible. Existing smart bins predominantly monitor a single parameter — fill level — and lack environmental and gas sensing capabilities.
ML-Based Prediction
Research demonstrates that machine learning models, particularly XGBoost regression, can effectively predict waste volumes using historical municipal records combined with rainfall, temporal, zonal, demographic, and calendar features. Dynamic, sensor-informed routing can reduce operational cost by up to 40% compared with static scheduling.
Smart Collection Systems
Existing solutions address waste collection with fixed schedules and limited real-time information. There is a need for an integrated mobile workflow connecting IoT-derived critical-bin information with resident–driver coordination, combining GPS tracking, route assistance, and proximity-based notifications.
Research Gap
Absence of Environmental Sensing
Most commercial smart bins report fill level only; hazardous-gas (methane, ammonia, CO₂) and thermal-environmental data are missing from the telemetry, leaving sanitation workers exposed to avoidable gas hazards.
Limited Behavioral Monitoring
Existing systems show that a bin is full but not how it got that way. There is no detection of illegal dumping, sudden fill spikes, abnormal lid activity, extended lid opening, or imminent overflow patterns.
Insufficient Data for ML Models
Single-parameter fill-level data provide insufficient features for waste composition prediction. Localized municipal-level forecasting combining environmental, temporal, and demographic features is largely unexplored for Sri Lankan contexts.
Weak Coordination Workflows
There is a gap in integrating IoT-derived critical-bin information with a unified mobile coordination workflow for resident–driver communication, live vehicle tracking, and proximity-based arrival notifications.
Research Problem & Solution
The Problem
Conventional municipal waste collection follows fixed routes and schedules with no real-time knowledge of bin conditions. Waste collection decisions are made without knowledge of the environmental state of each bin — a half-full bin emitting elevated methane or ammonia may need priority collection for safety and odour reasons, yet no fill-level-only system can flag it. This leads to:
- Premature collection of part-filled bins
- Delayed service to overflowing bins
- Higher fuel consumption and excess labour
- Worker exposure to hazardous gas conditions
- Avoidable environmental contamination
Our Solution
A comprehensive Smart Waste Management System that builds an end-to-end predictive framework combining four integrated components:
- IoT-based environmental and gas sensing with real-time cloud telemetry
- Smart fill-level monitoring with anomaly detection and behavior-aware tracking
- ML-based municipal waste prediction, composition estimation, and circular economy revenue modeling
- Mobile application for real-time bin detection, adaptive route optimization, and resident-driver coordination
Research Objectives
Environmental Sensing Smart Bin
Hemsara K.D.S. (IT22548528)- Design and build an ESP32-based sensing unit with MQ-135, MQ-4, and DHT22 sensors
- Implement GREEN/YELLOW/RED criticality classification using multi-parameter thresholds
- Publish timestamped JSON telemetry to Firebase Realtime Database
- Achieve defined targets for sensor accuracy, alert accuracy, delivery rate, latency and uptime
Smart Fill-Level & Occupancy Monitoring
Gamaariya C.C. (IT22202772)- Combine HC-SR04 ultrasonic sensor and magnetic reed switch on ESP32
- Classify bin fill into OK, Warning, and Critical status levels
- Implement rule-based anomaly detection for illegal dumping, fill spikes, and abnormal activity
- Achieve >95% fill-classification accuracy and >90% anomaly detection accuracy
ML-Based Waste Prediction & Revenue Hub
Jayawardhane K.G.M.S. (IT22213358)- Develop XGBoost regression pipelines for wet and dry waste prediction
- Integrate rainfall, temporal, zonal, demographic, and calendar features
- Build composition estimation, compactor planning, and resource-recovery modules
- Target >85% predictive accuracy on municipal test data
Mobile App for Real-Time Bin Detection, Adaptive Route Optimization & Resident-Driven Coordination
Wijesiri A.M. (IT22088482)- Integrate IoT-based smart-bin monitoring to identify bins reaching critical waste levels in real time
- Develop a mobile map interface that enables drivers to identify and prioritize critical bins
- Develop a resident–driver coordination mechanism for submitting and managing pickup requests
- Implement route support for drivers based on their current location and collection points
- Evaluate the usability, functionality, and effectiveness of the mobile-based system
Methodology
The project follows a design-science, design-and-development approach: each of the four components designs and builds a working artefact and evaluates it against predefined targets. Development proceeds through problem identification, solution design, artefact creation, and evaluation.
Problem Identification
Systematic literature review and gap analysis identifying limitations in current waste management approaches, covering IoT sensing, ML prediction, and mobile coordination.
Solution Design
Architecture design of four integrated components with defined interfaces. Hardware selection (ESP32, MQ sensors, DHT22, HC-SR04), ML model selection (XGBoost), and mobile framework decisions.
Artifact Development
Implementation of IoT firmware (Arduino/C++), cloud integration (Firebase Realtime Database), ML pipelines (Python/XGBoost) with a Flask back-end, web dashboards (React), and a Flutter mobile application with GPS tracking, live maps, and route support.
Evaluation & Validation
Testing against predefined metrics: sensor accuracy, fill-status classification, anomaly detection, prediction accuracy, cloud delivery and latency, and end-to-end functional test scenarios. See Results & Findings for the measured outcomes.
Technologies Used
Hardware & IoT
ESP32
Microcontroller
Cloud & Backend
Firebase
Realtime Database
Flask
ML Back-end (REST API)
MongoDB
Prediction Run Storage
FCM
Push Notifications
Cloud Firestore
Mobile App Data & Real-Time Streams
Firebase Auth
User Authentication
Cloudflare Workers
Serverless Backend (TypeScript)
ML & Data Science
Python
Programming Language
Scikit-learn
Data Processing
Frontend & Mobile
React
Web Dashboard
Flutter / Dart
Mobile App
Arduino/C++
Firmware
OpenStreetMap
Maps (flutter_map)
