Improving Suicide Ideation Detection Through Feature Engineering and Machine Learning

Suicide ideation detection from textual data is a challenging task due to the subtle and complex nature of language. This project explores the use of advanced machine learning and deep learning algorithms to identify text entries indicative of suicide ideation. The approach combines feature engineering techniques with models such as Convolutional Neural Networks (CNN), Bidirectional Long Short-Term Memory (BiLSTM), Bidirectional Gated Recurrent Units (BiGRU), BiLSTM with Logistic Regression, and RoBERTa with Logistic Regression. Text data is preprocessed to remove noise, normalize entries, and transform words into meaningful numerical representations. These steps enable the models to capture both semantic and sequential patterns, improving prediction accuracy. The system is deployed as a web application using Flask, providing a structured interface with modules for registration, login, text classification, and session management. Model performance is evaluated using accuracy, precision, recall, and F1-score to compare effectiveness. The comparative study demonstrates that hybrid models and pre-trained language models enhance detection capabilities over single-model approaches. This project provides a systematic methodology for detecting suicide ideation in textual data, highlighting the importance of feature engineering and model selection in improving classification outcomes.

Keywords: suicide ideation, text classification, machine learning, deep learning, feature engineering, CNN, BiLSTM, BiGRU, RoBERTa, Flask.