www.ijcrt.

org © 2024 IJCRT | Volume 12, Issue 7 July 2024 | ISSN: 2320-2882

Text-Based Stress Detection And Classification

Using Machine Learning
Harika Juluri1 , Deepender Singh2,T R Yograj Singh3,G Sai Kiran Yadav4
1Department of Computer Science and Engineering,
1Methodist College of Engineering and Technology, Hyderabad, Telangana, India

Abstract:As mental health awareness is increasing, there is a growing need for tools to help detect and
manage stress levels in individuals. It can be challenging for individuals to recognize and categorize their
stress levels accurately, leading to insufficient or inappropriate coping strategies. The aim of this work is to
develop a text-based stress detection and categorization model that utilizes natural language processing
techniques to analyze text messages and detect stress levels. This wok involves several steps, including
dataset preprocessing, feature extraction, training, stress detection, stress categorization and
recommendation generation. The model will be trained using advanced machine learning techniques and
evaluated using various performance metrics such as accuracy, precision, recall and F1 score. The results of
this project will enable users to manage their stress effectively.

Key words – Machine learning, Random Forest, XGboost, performance metrics

1.INTRODUCTION
1.1 Understanding Stress
Stress is ubiquitous and serious condition that exerts profound effects on both mental and physical
health. It arises when body responds to any change requiring adjustment or reaction, which can stem
from external factors like environmental pressures or internal factors such as personal thoughts and
emotions. The Pervasive ness of stress can lead to cascade of emotional, physical, and behavioral
isssues, significantly disrupting and individual’s daily life and overall well-being. The biological
response to stress involves the activation of the hypothalamic-pituitary-adrenal (HPA) axis, leading to
the release of cortisol and other stress hormones, which prepare the body for a fight-or-flight response.
However, chronic activation of this response lead to detrimental health outcomes, including
cardiovascular disease, weakened immune function and mental health disorders like anxiety and
depression.

1.2 Symptoms and Impact

Stress manifests in various symptoms that can significantly impact a person’s mental and physical
health. Emotional symptoms include irritability, moodiness and feeling overwhelmed, while physical
symptoms can range from low energy and headaches to rapid heartbeat and chest pain. Behavioral
changes such as changes in appetite, procrastination and increased use of substances like alcohol or
drugs are also common. These symptoms highlight the extensive reach fo stress throughout the body
and underscore the need for effective detection and management strategies.

1.3 Addressing stress through technology

The development of tools and strategies to detect and address stress is an ongoing area of research and
innovation, aiming to provide individuals with the resources they need to maintain mental and physical
health. Text-based stress detection and classification using machine learning represents a promising
approach in this field. By analyzing linguistic patterns and emotional cues in written text, ML
Algorithms can identify indicators of stress, enabling timely interventions and support. This project
IJCRT2407927 International Journal of Creative Research Thoughts (IJCRT) www.ijcrt.org i282
www.ijcrt.org © 2024 IJCRT | Volume 12, Issue 7 July 2024 | ISSN: 2320-2882

explores the potential of ML in stress detection, contributing to the broader efforts to improve mental
health outcomes through technological advancements.

2. Literature Survey

2.1 Paper Title: “A Novel approach for text classification by combinig pre-trained BERT model with
cnn classifier”
Authors: Chenxu Wang, Yulin Li, Ziying Wang.

Key findings:
 The proposed model which combines a pre-trained BERT model with a CNN classifier, achieves
the highest performance on the AG News and Amazon product reviews datasets compared to
baseline models.
 The model’s superior performance is attributed to its ability to leverage the contextualized
embeddings of BERT and the n-gram feature capturing of CNN.

2.2 Pape Title: “Research on domain text classification on method based on BERT”
Authors: Shaohui Xie, Yangsen Zhang, Zhengyu Hou, Zhenjiang Su, Renjie Wang, Ziyuan He.

Key findings:
 The proposed BERT_VCA Model outperformed the BERT_BASE modle by 1.12% in F1 value
for domain text classification.

2.3 Paper Title: “Stress detection using natural language processing and machine learning over social
interactions:
Authors: Tanya Nijhawan, Girija Attigeri, T Anantha Krishna

Key findings:
 Random forest classifier achieved the highest accuracy of 97.78% among the machine learning
models test for sentiment analysis.
 The BERT Deep learning model achieved an accuracy of 94% for emotion classification.

2.4 Paper Title: “Bert-Enchanced Text Graph Neural Network for Classification”
Authors: Yiping Yang, Xiaohui Cui

Key findings:
 BEGNN which combines BERT-based and GNN-based feature extraction with a co-attention
module, outperforms other baseline models.
 BEGNN shows more obvious imporvements on datasets with longer text, indicating it can better
process longer texts.

2.5 Paper Title: “How to fine-tuen BERT for Text Classification?”

Authors: Chi Sun, Xipeng Qui, Yige Xu, Xuanjing Huang.

Key findings:
 The top layer of BERT is the most useful for the text classification tasks.
 Using an appropriate layer-wise decreasing learning rate helps BERT overcome the catastrophic
forgetting problem.

IJCRT2407927 International Journal of Creative Research Thoughts (IJCRT) www.ijcrt.org i283

www.ijcrt.org © 2024 IJCRT | Volume 12, Issue 7 July 2024 | ISSN: 2320-2882

3. Proposed Work

3.1 Introduction

Briefly introduce the problem of stress among the people and the impact of stress on individual’s health.
Highlight the importance of early detection of stress and intervention. State the research objectives and the
significance of the proposed work.

In today’s world, many of us rely on social media platforms such as Facebook, X (formerly Twitter),
Snapchat, YouTube, TikTok, and Instagram to find and connect with each other. While each has its
benefits, it’s important to remember that social media can never be a replacement for real-world human
connection. Human beings are social creatures. We need the companionship of others to thrive in life, and
the strength of our connections has a huge impact on our mental health and happiness. Being socially
connected to others can ease stress, anxiety, and depression, boost self-worth, provide comfort and joy,
prevent loneliness, and even add years to your life. On the flip side, lacking strong social connections can
pose a serious risk to your mental and emotional health.

3.2 Literature Review

Summarize existing research on stress detection, including methodologies, datasets and outcomes. Identify
the key findings and gaps in the current literature. Discuss the application of machine learning in the
context of veterinary epidemiology.

3.2 Research objectives

Clearly define the main goals of the proposed research. For Example: To develop a machine learning
model for detection of stress among individuals using text input of their messages or posts. To create a
user-friendly interface for individuals to post their feelings and use the model.

Detection of stress in texts from social media by current systems is quite a challenge. This is because
models rely on the sensory data along with the text messages to predict stress, but in all the scenarios the
user may not have access to such IoT devices or sensors to check stress making the models less accessible
and user friendly.

3.3 Methodology

Explain the data collection process, including sources and types of data. Describe the preprocessing steps,
data cleaning and feature engineering. Discuss the selection of machine learning algorithms and
techniques( e.g., Decision Trees, Support Vector Machine, Neural Networks). Explain the model
evaluation and validation process.

3.4 Data collection and Preprocessing

Detail how data was collected, anonymized and processed. Address issues related to data quality, missing
values and data imbalance. Discuss the ethical consideration reagarding data usage and privacy.

Word2Vec is a powerful technique for natural language processing that captures semantic
relationships between words. Developed by researchers at Google in 2013, Word2Vec transforms text into
a numerical form that algorithms can easily process. This model creates word embeddings, which are
dense vectors representing words in a continuous vector space, where words with similar meanings are
positioned close to each other. The fundamental concept behind Word2Vec is to use the context of a word
within a sentence to predict the word itself or vice versa.

BERT (Bidirectional Encoder Representations from Transformers) tokenizer is an advanced natural

language processing tool designed to pre-process text data for the BERT model, a state-of-the-art
transformer-based architecture developed by Google in 2018. The BERT tokenizer breaks down text into
tokens, which are the basic units of meaning in the model. Unlike traditional tokenizers that often split text
into words, the BERT tokenizer employs a WordPiece algorithm that breaks words into subword units.
This allows BERT to handle out-of-vocabulary words effectively and understand context more precisely
IJCRT2407927 International Journal of Creative Research Thoughts (IJCRT) www.ijcrt.org i284
www.ijcrt.org © 2024 IJCRT | Volume 12, Issue 7 July 2024 | ISSN: 2320-2882

by considering the subwords that make up a word. For instance, the word "unhappiness" might be
tokenized into ["un", "##happiness"], allowing BERT to understand its components and contextual
meaning more accurately.

3.5 Model Development and Training

Explain the development of machine learning models for the detection and classification of stress.
Describe the algorithm selection, cross-validation strategies. Discuss the choice of evaluation metrics and
model interpretation techniques.

The Random Forest Classifier is a robust and versatile machine learning algorithm renowned for its
exceptional performance in classification tasks. It operates by creating an ensemble of decision trees,
where each tree is trained on a random subset of the data and features, ensuring diversity and reducing
overfitting. This ensemble approach aggregates the predictions of multiple decision trees, resulting in a
highly accurate and stable model.trees, resulting in a highly accurate and stable model. Random Forest
Classifier is particularly effective in handling complex datasets with both numerical and categorical
features, making it suitable for a wide range of applications. Its ability to assess feature importance
provides valuable insights into the underlying data patterns, aiding in feature selection and model
interpretation. Additionally, Random Forest Classifier is scalable and computationally efficient, capable of
handling large datasets with high dimensionality. Its robustness to noisy data and flexibility in handling
various data types make it a popular choice among data scientists and machine learning practitioners for
building reliable and accurate classification models.

XGBoost (Extreme Gradient Boosting) is a powerful machine learning algorithm known for its
exceptional performance in a wide range of tasks, including regression, classification, and ranking. It
belongs to the family of boosting algorithms, which combine weak learners (typically decision trees) to
create a strong predictive model. XGBoost has gained popularity due to its scalability, flexibility, and
ability to handle complex datasets. The key idea behind XGBoost is to iteratively build an ensemble of
weak prediction models, called decision trees, to improve the overall predictive performance. The
algorithm consists of three main components: a loss function, a weak learner, and a boosting framework. It
is a powerful machine learning algorithm that combines the predictions of multiple weak learners to create
a strong predictive model

3.6 User Interface Development

If your research includes creating a user-friendly tool outline the development process. Mention the
technologies and frameworks you plan to use for the interface

3.7 Results

Present the expected outcomes, such as the accuracy of the stress detection and classification model
respectively. Model performance, the usability of the developed tool.

4 System Architecture

IJCRT2407927 International Journal of Creative Research Thoughts (IJCRT) www.ijcrt.org i285

www.ijcrt.org © 2024 IJCRT | Volume 12, Issue 7 July 2024 | ISSN: 2320-2882

5. Conclusion

Text-based stress detection using machine learning holds immense promise for transforming the way we
manage mental health and welfare. By harnessing the power of data, artificial intelligence and predictive
modelling, we can work towards a future where stress detection, prevention an effective management
become the norm in various contexts. This innovative approach offers the potential to improve stress-
management outcome.

6. References

[1]Shaohui Xie, Yangsen Zhang, Zhengyu Hou, Zhenjiang Su, Renjie Wang, Ziyuan He – “Research on
domain text classification method based on BERT”

[2]Chenxu Wang, Yulin Li, Ziying Wang –“A Novel Approach for Text Classification by Combining Pre
trained BERT Model with CNN Classifier”

[3] Tanmay Gupta, Samarth Singh, Mr Arun Kumar – “ Community Conversation Analyser”

[4] Ravishankara K, Dhanush, Vaishak, Srajan – “whatsapp chat Analyser”.

[5] Weiguo Fan, Michael D Gordan – “The Power Of Social Media Analytics”

[6] M Ikonomakis, S Kotsaintis, V Tampakasm – “Text classification using machine learning techniques”

[7] Kern M L, Park G, Echstaedt J C, Schwartz H A, Sap M, Smith L K, Ungar L H – “Gaining insights
from social media language: Methodologies and Challenges”.

[8] Sergio Munoz, Carlos A Iglesias – “Information Processing and Management”.

[9] Andrian Benton, Margaret Mitchell, Dirk Hovy – “Multi-task learning for mental health using social
media texts”.

[10] Joseph Lilleberg, Yun Zhu, Yanqing Zang – “support vector machine and wordevec for text
classification with semantic features”.

[11] M Nii, Yuya Tuchida, T Iwamoto, A Uchinuno, R Sakasita – “Nursing-care text evaluation using
word vector representations realized by word2vec”

[12] Yonghui Zhang, Jingang Liu – “Microblogging Short Text Classification Based on Word2Vec”

[13] Lei Zhu Guijun Wang, X. Zou – “A Study of Chinese Document Representation and Classification
with Word2vec”

[14] Md. Rafqul Islam, Ashir Ahmed,Abu Raihan M. Kamal,HuaWang,Anwaar Ulhaq - “Depression
detection from social network data using machine learning techniques”

[15] Jeremy Howard, Sebastian Ruder – “Universal Language Model Fine-tuning for Text Classification”.

[16] Chi sun, Xipeng Qiu, Yige Xu, Xuanjing Huang – “How to Fine-Tune BERT for Text
Classification?”.

IJCRT2407927 International Journal of Creative Research Thoughts (IJCRT) www.ijcrt.org i286