A Deep Gated Residual Auxiliary Multilayer Perceptron Model for Early Detection of Malnutrition in Under-Five Children in Rural Tamil Nadu

Abstract

Malnutrition among children remains a primary public health issue in India, with tangled demographic, socio-economic and clinical factors leading to malnutrition. The future generation is fundamentally dependent on the healthy childhood. The healthy child hood provides the foundation for cognitive development, emotional resilience and physical growth. Shaping these early phase of the life bring values, aspiration and capabilities in individual. It is one of the strategies to build more evolved, equitable and sustainable future. Tamil Nadu performs better than other states of India in nutritional programs for children under 5 years. The current programs like Uttachathai Uruthi Sei, breakfast schemes and RBSK screenings are showing nutrition security and community-based behavioural changes in Tamil Nadu. Still it is challenging to implement these programs in rural, tribal and urban poor areas. The two phases of screenings were conducted under the scheme of Rashtriya Bal Swasthya Karyakram (RBSK). The first phase (Apr – Sep 2024) reveals that 10.2% children are underweight, 3.3% are affected by Moderate Acute Malnutrition (MAM) and 1.9% children are suffered by Severe Acute Malnutrition (SAM). During the second phase (Oct – Mar 2025) 8% children are underweight, 3.8% are MAM and 2.5% are SAM. Though there is a marginal reduction in overall malnutrition from the first phase (15.4%) to the second phase (14.5%), the individual attention is needed to rectify this problem. The traditional screening techniques are relying on anthropometric measurements. The multi-dimensional analysis is required to solve an each individual child’s malnutrition. This study presents the novel deep learning model - Deep Gated Residual Auxiliary MLP Model (D-GRAM) for early malnutrition detection in children under-five years. The multi-dimensional features dataset were collected from Anganwadi centres and mothers of the children from Thirumanur Union under the Ariyalur District, Tamil Nadu. This proposed model was developed with several optimization techniques and achieved 98% accuracy.