Soham Ghosal

Patient Finding Analytics - Type 1 Diabetes Prediction

• Objective: Identify individuals at the risk of Diabetes (Type 1), facilitating early intervention and lowering the risk of death over a span of 15-20 years based on clinical and lifestyle factors.
• Methodology: Design an API-based screening tool for real-time patient risk assessments using Random Forest model, identifying 500+ eligible individuals weekly. Optimized data handling with preprocessing techniques, improving workflow efficiency and increasing processing speed by three hours per week.
• Business Impact & Applications: Improved patient treatment success by 30%, driving 5-7% revenue growth and market expansion..

Research Project - Synthetic Data Generation

• Objective: Build an AI-driven solutions that replicate real-world data while preserving patient privacy.
• Methodology: Developed an innovative approach with GANs to create varied medical record datasets increasing efficiency within data processing cycles; established benchmarks resulting from model testing that achieved consistent accuracy rates above industry standards at over 92%.
• Business Impact & Applications: Reduces reliance on costly patient surveys, lowering data collection costs by 15-20%, and accelerates innovation 3x for faster market deployment.

Classification Modeling - Crop Classification

• Objective: Automate crop classification using machine learning models and statistical analysis, decreasing manual effort, minimizing errors, and boosting accuracy and efficiency.
• Methodology: Crafted a Random Forest model utilizing vegetation indices, attaining an F1 score of 86% across key crop classifications; this methodology now supports agricultural banks in processing loan applications for farmers.
• Business Impact & Applications: Increased company revenue by 10-12% through improved reporting accuracy, partnered with the Uttar Pradesh government to develop an automated crop monitoring system benefiting 1M+ farmers, and secured $500K+ in annual funding for smallholder farming initiatives.

Statistical Modelling - Irrigation Mapping & Cropping Intensity

• Objective: To create an unsupervised pixel-wise classification model to identify irrigated and rainfed areas, leading to accurate mapping of 250+ agricultural zones using Sentinel2 satellite imagery for enhanced yield estimation.
• Methodology: Analyzed 2,000+ satellite images to implement mathematical and statistical models for spectral indices; increased classification accuracy of agricultural activity across diverse geographical locations, directly impacting farm report reliability.
• Business Impact & Applications: Both the products are important parameters in the farm report which contributes around 40% of company revenue.

Regression Modeling - Soil Moisture Estimation

• Objective: Build a model to estimate soil moisture levels based on environmental and soil parameters, enabling efficient water resource management and precision agriculture.
• Methodology: Analyzed historical climate data alongside current conditions using advanced statistical models which predicted variation in soil moisture within 5% accuracy margins across different field locations throughout the growing season.
• Business Impact & Applications: Optimized irrigation using soil moisture data, reducing water waste by 25% and boosting harvest yields, while securing partnerships with four key agricultural stakeholders to enhance operational efficiency.