BITI 1113 Artificial Intelligence Project Report 1

Revolutionizing Waste Sorting: A Machine

Learning for Recyclable or Non-Recyclable
A. Nuar , N. A. Azli , A. M. Nizam , S. F. S Ja’afar .

To address these challenges, there is a growing interest in

leveraging advanced technologies to automate and improve the
Abstract - Effective waste management is critical in recycling process. Machine learning and artificial intelligence
addressing global environmental challenges, particularly offer promising solutions by enabling automated sorting and
through enhanced recycling practices. This study explores classification of waste. These technologies can significantly
the application of machine learning and artificial enhance the accuracy and speed of recycling operations,
intelligence (AI) for classifying waste into recyclable and reducing the reliance on manual labor, and minimizing errors.
non-recyclable categories. Conventional recycling methods By automating the sorting process, recycling centers can
often suffer from inefficiencies and contamination due to operate more efficiently, handle larger volumes of waste, and
manual sorting processes. Leveraging AI technologies achieve higher recycling rates.
promises to automate and improve sorting accuracy,
thereby enhancing overall recycling effectiveness. This Furthermore, the integration of AI in waste management
paper presents a classification model using the Teachable supports the development of intelligent systems that can
Machine platform, trained on a diverse dataset of waste provide real-time insights and feedback. Such systems can help
item images. The model enables real-time feedback on the in better tracking and managing waste streams, optimizing
recyclability of items uploaded by users, promoting recycling efforts, and promoting sustainable practices. The use
sustainable waste management practices. The of data analytics and machine learning models can facilitate
implementation demonstrates significant potential to more informed decision-making, allowing for proactive
streamline recycling operations, reduce errors, and foster measures to improve waste management operations.
environmental sustainability.

Index Terms—Recyclable, non-recyclable, machine learning,

artificial intelligence, Teachable Machine II. FRAMEWORK OF RECYCLABLE OR NON-
RECYCLABLE CLASSIFICATION.
I. INTRODUCTION
The framework for classifying waste as recyclable or non-
The increasing amount of waste produced due to recyclable using a Teachable Machine revolves around several
urbanization and population increase makes waste management interconnected pillars aimed at improving waste management
and recycling an urgent global concern. [1] The United Nations practices and promoting environmental sustainability. Firstly,
Environment Program states that more than two billion tons of the system involves the collection of a diverse dataset of waste
municipal solid trash are produced annually worldwide. item images, meticulously labeled as recyclable or non-
Conventional recycling facilities are finding it increasingly recyclable. This dataset is the foundation for training a
difficult to manage this increase since they mostly rely on machine-learning model using the Teachable Machine
human sorting procedures. These facilities use human laborers platform. The trained model is deployed onto the web, enabling
to separate materials like glass, aluminum, paper, and plastic users to upload images of waste items and receive real-time
which is labor-intensive, sluggish, and prone to human mistake feedback on their recyclability status.
process.
Additionally, the framework emphasizes the integration of
The inefficiencies of manual sorting cause increased rates of sustainable waste management practices, advocating for
contamination of recyclable materials in addition to slowing responsible recycling behavior and minimizing contamination
down the recycling process. Contaminated recyclable materials in recycling streams. It supports educational initiatives to raise
frequently lose their ability to be recycled, lowering overall public awareness about proper waste segregation and the
recycling rates, and increasing the quantity of waste that ends benefits of recycling. Furthermore, the framework promotes
up in landfills. The urgent need for waste management collaboration with waste management services to streamline
techniques to innovate is highlighted by this inefficiency.
waste collection and processing, leveraging data analytics to
optimize recycling efforts and track progress toward
environmental goals. Continuous evaluation and improvement
BITI 1113 Artificial Intelligence Project Report 2

are integral to ensuring the model's accuracy and effectiveness and helpful in teaching them about proper waste management
in diverse operational settings. Overall, this framework practices.
harnesses technology to enhance waste classification accuracy,
promote recycling, and contribute to a more sustainable future. Finally, the proposed predictive model effectively improves
recycling processes and promotes sustainable practices. Future
improvements may include expanding the data set to include a
wider range of items, integrating additional features such as
III. THE PROPOSED PREDICTIVE MODELLING
accessibility features, and partnering with recycling centres to
implement the system on a larger scale. Artificial intelligence
An effective method for streamlining recycling procedures is and predictive modelling can revolutionize recycling practices,
predictive modelling. Using the Teachable Machine, we aim to improve efficiency rates, and make significant steps in
create an accurate model that can classify recyclable and non- promoting sustainable waste management.
recyclable products through image recognition. This approach
not only simplifies recycling procedures but also increases
efficiency and promotes sustainable waste management IV. PROJECT DESCRIPTION
practices.
The purpose of our project is to make recycling easier and
To train the predictive model, we collected a large dataset of better. Currently, sorting recyclables like plastic and paper can
images representing various recyclable and non-recyclable take a long time, and mistakes can sometimes happen. So, we're
materials such as plastic, paper, aluminum, glass, rubber, using machine-learning technology to help with sorting.
clothes, kitchen waste and diapers and pet waste. These images Our main goal is to teach computers to recognize if something
were sourced from public databases, online repositories, and is recyclable or not just by looking at it. We want to ensure that
actual samples, each labelled to indicate whether or not they are things like plastic bottles and aluminum cans are in the right
recyclable. To ensure the accuracy of the training, high-quality place. But our project goes beyond just sorting. We also want
images were selected that clearly showed the items. to help stop contamination and educate people about recycling.
By using this technology, we hope to make recycling more
The pre-processing of data included resizing all images to efficient and encourage everyone to do their part for the
uniform dimensions, normalizing pixel values and applying environment.
data augmentation techniques such as rotation, translation and
cropping to increase the diversity of the training and improve Overall, our project aims to revolutionize how we recycle,
the generalization of the model. This prepared the images for making it easier and more accessible for everyone.
efficient training by the Teachable Machine.

Model training was performed using Teachable Machine's V. RESULT AND DISCUSSION
web-based user interface. The training involved loading the
images into preprocessed classes and dividing them into For the accuracy test, we used the image to identify the types
recyclable and non-recyclable classes, as well as setting training of waste through image recognition. The accuracy test revealed
parameters such as number of epochs and learning rate. Real- that our model could achieve an accuracy rate typically around
time feedback during training helped monitor model 89.5%, indicating a strong performance in correctly classifying
performance and make necessary adjustments. recyclable and non-recyclable items. This high accuracy rate is
attributed to the balanced dataset and thorough preprocessing
After the training process was completed, the model was techniques applied during the training phase.
evaluated using metrics such as accuracy, precision, recall and
confusion matrix to evaluate its performance in identifying We collected a total of 322 images, evenly distributed among
recyclable and non-recyclable items. The trained model was various categories of recyclable and non-recyclable items.
then integrated into a user-friendly interface that allows users Specifically, we uploaded more than 20 images each for plastic,
to upload images, receive immediate feedback on the paper, aluminum, glass, rubber, and clothes under the
recyclability of items, and receive guidance on appropriate recyclable category. For the non-recyclable category, we
disposal practices based on classification results. included more than 20 images each for kitchen waste and
diapers and pet waste. The balanced dataset helped ensure that
The integrated system has been tested on a separate the model could learn to differentiate between each type of
validation dataset to ensure its reliability and accuracy in real- waste effectively.
world scenarios. This involved real-world testing with images
taken in different lighting conditions and backgrounds and The following results were established during the testing
gathering user feedback to identify usability issues and areas for phase using images that were outside of samples in Teachable
improvement. The predictive model showed high accuracy, Machine.
which greatly improved sorting efficiency and reduced
classification errors. Users found the interface to be intuitive
BITI 1113 Artificial Intelligence Project Report 3

Item Output/Result
Paper 100%
Aluminum 100%
Clothes 59%
Rubber 95%
Glass 73%
Plastic 96%
Kitchen waste 93%
Diapers & Pet waste 100%

Kitchen Waste

Testing:

Paper

Aluminum

Plastic

Clothes

Rubber

Strength:
Diaper & Pet Waste i. Ease of use
Teachable Machine offers an exceptionally user-friendly online
platform that simplifies the process of training machine
learning models, eliminating the need for extensive technical
knowledge. Additionally, its web-based nature allows for easy
access from any location with an internet connection, making it
perfect for collaborative efforts.

ii. Integration Capabilities

Teachable Machine seamlessly integrates with web
applications, facilitating the creation of web-based recyclable
Glass or non-recyclable classification solutions that are easily
BITI 1113 Artificial Intelligence Project Report 4

accessible to a wide audience. The platform’s support for VII. ACKNOWLEDGMENT

various data formats, including images, audio, and poses,
We would really like to express our sincere gratitude to all
broadens its applicability beyond waste classification, enabling
those who supported and guided us throughout this project.
its use in a wide range of machine-learning projects.
First and foremost, we would like to thank our lecturer, Dr
Halizah binti Basiron, for her helpful guidance, insightful
Challenges:
comments, and non-stop encouragement, which have been
i. Limited Model Complexity
instrumental in the successful completion of this project. We
Teachable Machine primarily supports simple machine learning
amplify our appreciation to our group members, Aziah binti
models, which may not be sufficient for complex waste
Nuar, Nurathirah Aina binti Azli, Nur Anis Maisarah binti
classification tasks that require deep learning or advanced
Mohd Nizam and Siti Fatimah Safiqah binti Ja'afar, for the hard
neural network architectures. The basic models may also
work, dedication, and collaborative spirit. Each member's
struggle to maintain high classification accuracy.
particular contributions and dedication were important in
growing our machine learning model using the Teachable
ii. Inaccuracy Due to Limited Training Data
Machine. We are also thankful to Google for supplying the
It tends to produce inaccurate models if trained on a limited
Teachable Machine platform, a user-pleasant and effective
amount of data. Small datasets often fail to capture the diversity
device that enabled us to create and educate our version with
and complexity of waste materials, leading to poor
ease. The accessibility and performance of this platform played
generalization and classification performance.
a vital position in our project's success. Special thanks to our
friends and classmates who provided valuable comments and
How to Improve:
suggestions throughout the checking out phase of our model.
i. Enhance Data Collection
Their insights helped us refine and enhance the accuracy and
Increase the variety and quantity of waste item images,
usefulness of our system. Finally, we would like to
including rare and complex waste types, to improve model
acknowledge the support of our families and friends, whose
robustness and accuracy. This can be achieved by sourcing data
encouragement and understanding helped us stay focused and
from diverse environments and conditions.
motivated during this journey. Thank you all for your support
and contribution to the successful completion of our project.
ii. Improve Scalability and Performance
Enhance the platform’s ability to handle larger datasets and
higher computational loads, ensuring smooth performance for
more extensive and complex projects.
VII. REFERENCE
[1] Global Waste Management Outlook 2024.(n.d.b).UNEP -
UN Environment Programme.
VI. CONCLUSION https://www.unep.org/resources/global-waste-management-
outlook-2024
This project tackled the important issue of recyclable and
non-recyclable management. By using the Teachable Machine
platform, we created a system that can accurately tell if waste
items are recyclable or not through image recognition. This
system makes sorting faster and reduces mistakes, leading to
better recycling practices and more efficiency. Our model
showed a high accuracy rate of 89.5% in identifying waste
materials, thanks to a well-prepared dataset and effective
training methods. Real-world tests confirmed the model’s
reliability in various conditions and for different types of waste.
While the project was successful, we faced challenges like
limited data and model complexity. To improve, we need to
gather more diverse waste images and enhance the platform’s
ability to handle larger datasets. Continuous updates and
feedback will help keep the model effective. The user-friendly
Teachable Machine platform made it easy to integrate our
solution into recycling processes, allowing for broader adoption
and better waste sorting practices. In summary, our project
shows how technology can improve recycling by making it
more accurate and accessible. With further development, we
aim to significantly enhance waste management and contribute
to a cleaner environment.