Stress and Depression Detection

About the stress and depression detection M.Tech project

Many students go through long stretches of stress or low mood without telling anyone, yet writing a few lines in a journal comes easily to most of them. This stress and depression detection M.Tech project turns that habit into an early, private signal. A student writes a daily entry in an Android app, a language model reads it, and only when worrying signs repeat over several days does the app suggest the DASS-21 self-test and a counsellor session.

The app is a screening aid, not a diagnosis. That boundary is part of the design and gives your thesis a clear ethics section.

Research behind it

Base paper. The study builds on Calibration of Transformer-Based Models for Identifying Stress and Depression in Social Media, published in IEEE Transactions on Computational Social Systems (2024). It joins extra-linguistic features with BERT and MentalBERT through a gate, uses label smoothing so that confidence matches correctness, and tests on three public social media datasets.

Research gap. Social media posts are public, short and written for an audience. Journal entries are private and first-person, so a model trained only on posts may misread them. Each post is also judged alone, while a student's state shows over days. Finally, when a confidence value triggers an alert to a counsellor, calibration has a direct cost in false alarms.

Proposed system and novelty. MentalBERT first continues its masked language training on diary-style sentences, so it learns how people write about their own day. It is then fine-tuned on Dreaddit and a Reddit depression dataset. Temperature scaling, compared with label smoothing, calibrates the output. Each new entry updates a weighted weekly score, and an alert fires only when that score stays high. A counsellor chosen by the student sees highlighted phrases behind each result.

Models compared. TF-IDF with SVM, a BiLSTM, DistilBERT and the base paper's gated MentalBERT re-implemented, against the proposed model, on public test sets and a journal-style test set. The ablation drops domain adaptation, swaps the calibration method and compares single entries with a multi-day window.

Sentiment at a finer level is the focus of the aspect based sentiment analysis project, if you prefer an NLP topic outside healthcare.

How it works

• The student opens the app with a PIN and writes the day's entry.
• The entry goes to the Flask API, where the calibrated model scores it.
• The score updates the weighted weekly trend on the phone.
• If the trend stays above the limit, the app suggests the DASS-21 self-test.
• The student can share the alert and ask for a counsellor session.
• The counsellor reviews the trend and phrases, adds notes and books a meeting.

Users and modules

There are three roles. The student keeps the journal, sees mood trends, takes the self-test and downloads a summary. The counsellor sees only students who chose to share, reviews charts and phrases, and records session notes. The admin manages accounts and records, checks the dashboard and model comparison page, and sets the window, alert limit and helpline numbers. Journal text stays private unless the student shares it.

What you get

• Phase 1 Research: Base paper study: calibrated transformers for stress and depression (IEEE TCSS, 2024); Literature survey of 15-20 recent mental health NLP papers; Research gap, novelty and objectives; Dreaddit and depression data EDA notebook; Synopsis and Phase-1 report.
• Phase 2 Implementation: Android journal app (Java); Flask API, counsellor and admin panels; Baselines and base model re-implementation; Domain-adapted, calibrated MentalBERT; Multi-day aggregation and alert logic; MySQL database and trained models.
• Results: Macro F1 and recall per class; Calibration error and reliability diagrams; False alerts per week and ablation graphs; results.xlsx and all graphs in high resolution.
• Thesis and paper: Thesis written for you (about 90 pages); IEEE conference paper draft; Journal paper draft; Publication guide.
• Reviews, viva and support: PPTs for Review 1, Review 2 and final viva; Viva book with 50+ questions; A1 poster and project diary; Demo and code walkthrough videos; 1-on-1 setup, 2 revisions and support till viva.

Who can use this project

The project suits M.Tech and M.E. students of CSE and IT looking for an NLP dissertation with a social purpose. If you searched for a MentalBERT text classification project, a model calibration temperature scaling project or a mental health NLP Android project, this covers all three, with an app your guide can try live.