This repository contains the implementation of the project Utilizing Machine Learning Models to Classify Suicidal Thoughts Using Data from Twitter/X. The project aims to detect suicidal tendencies based on data from the Twitter/X social media platform using various Natural Language Processing (NLP) techniques and Machine Learning models.
Every year, approximately 703,000 people worldwide end their lives, making suicide the fourth leading cause of death among individuals aged 15-29 years. In Indonesia, 6,544 suicide-related deaths were recorded in 2019. By leveraging data from Twitter/X, this project aims to enable early detection of suicidal thoughts and direct individuals to support services such as suicide hotlines or psychologists.
The dataset was collected by scraping Twitter/X using Python. Keywords like "capek hidup" (tired of life), "depresi" (depression), and "nyerah" (give up) were used to gather approximately 3,000 tweets in CSV format.
The preprocessing steps included cleaning and transforming the data:
- Data Cleaning: Removing irrelevant elements like mentions, hashtags, and URLs.
- Normalization: Converting informal or abbreviated words into their formal forms.
- Spell Correction: Fixing spelling errors.
- Stopword Deletion: Removing insignificant words.
- Tokenization & Lemmatization: Splitting text into smaller units and converting them into their root forms.
- Labeling: Using the DistilBERT model to label tweets as neutral, positive, or negative.
The models used were:
- Random Forest 🌳
- Support Vector Machine (SVM) 📈
- IndoBERT 📚
The models were tested to achieve the following accuracies:
- SVM: 84.8%
- Random Forest: 84%
- IndoBERT: 87%
The process and experiments can be accessed in this Repository
The IndoBERT model outperformed others with an accuracy of 87%, demonstrating its superior ability to understand complex textual contexts.
- Focuses on mental health issues for social media users.
- Uses original data.
- Implements state-of-the-art NLP models like IndoBERT.
- The dataset consists of only 3,000 tweets, causing the model to be biased towards negative data.
- No deployment as a web application has been implemented yet.
This project demonstrates that the IndoBERT method is effective in detecting suicidal tendencies in tweets. With additional data and better dataset balancing, the model's accuracy could be further improved.