A Hybrid Deep Learning Framework for Personalized Women’s Nutrition Recommendation Based on Menstrual and Health Parameters

Abstract: This research proposes a comprehensive framework for recommending personalized nutritional supplements for women, focusing on menstrual health and individual physiological parameters. The system begins with input data comprising menstrual symptoms (such as mood swings, fatigue, and cramps) and individual health indicators (such as age, BMI, and stress levels). Pre-processing techniques like missing value imputation, Z-score normalization, and one- hot encoding are applied to prepare the data. Feature extraction is carried out using a Layered Sparse Autoencoder Network, capturing complex patterns in the input. These features are then fed into a Hybrid Attention-based Bidirectional Convolutional Greylag Goose Gated Recurrent Network, which predicts nutrition recommendations. The model's performance is optimized using the Greylag Goose Optimization Algorithm. The final output suggests appropriate levels (High/Medium/Low) of essential nutrients such as Magnesium, Iron, Calcium, and Vitamin D, offering a data-driven and adaptive solution to women’s nutritional health.

Keywords: BMI, Z-score normalization, one-hot encoding, GGO, RNN, CNN, RNU