HairSentinel: a time-aware anomaly detection framework for forecasting hairfall trends using temporal fusion...

Area of Science:

• Computational biology and bioinformatics
• Health informatics
• Data science

Background:

• Hairfall is a widespread concern influenced by genetics, scalp health, nutrition, hormones, and sleep.
• Traditional hairfall detection methods rely on complex image analysis (CNN, SVM), limiting accessibility.
• A need exists for simpler, user-centric approaches to monitor hairfall patterns over time.

Purpose of the Study:

• To develop and validate a novel, user-friendly approach for detecting hairfall trends using time-series data.
• To compare the efficacy of different time-series anomaly detection models for hairfall analysis.
• To establish a system for proactive identification of potential health risks associated with hairfall fluctuations.

Main Methods:

• User-provided data collected via simple, time-centric questions (daily/weekly).
• Time-series anomaly detection using LSTM, Random Forest, Temporal Fusion Transformer (TFT), and ARIMAX models.
• Comparative analysis of models based on various performance metrics.

Main Results:

• The Temporal Fusion Transformer (TFT) model demonstrated superior performance.
• TFT achieved 97.5% accuracy and 97.4% precision in anomaly detection for hairfall.
• The study successfully established normal deviation margins for hair shedding cycles.

Conclusions:

• The proposed user-friendly method effectively models hairfall fluctuations.
• TFT is the most suitable model for proactive anomaly detection in hairfall data.
• This approach facilitates early identification of health risks, such as hormonal imbalances, and supports personalized dietary recommendations.