In today's world, women have made significant strides across diverse fields, yet the persisting issue of malnutrition among them poses multifaceted challenges. Beyond the immediate health implications, malnutrition in women exacts a heavy toll on the financial stability of families, the vitality of communities, and the overall prosperity of nations. Improved nutrition doesn't just enhance their physical health; it also bolsters their ability to contribute effectively to the workforce, igniting an upward trajectory for entire generations. Breastfeeding women are paramount. They require an additional 50% more calories than usual and a diet rich in balanced nutrition. In India, the challenge of malnutrition persists partly due to the absence of standardized methods to forecast and preempt it before it escalates. With limited access to medical facilities in many regions, by recognizing these patterns beforehand, targeted preventive measures can be implemented, potentially averting the escalation of malnutrition-related issues. This paper proposes a two staged LSTM based deep learning approach called Improved LSTM (i-LSTM) for predicting the occurrence of malnutritional defects by analyzing the health pattern over the past history in pregnant women. The proposed prediction system will be very beneficial for tribal pregnant women to undertake preventive measures and clinicians can easily diagnose the risks of pregnancy complications easily.