Yatharth Mudgil

Project 1 : Melanoma Detection using CNN - Convolutional Neural Networks | Image Classification | TensorFlow | Healthcare AI

• Developed a deep learning model to classify skin lesion images for early melanoma detection, addressing a critical healthcare challenge that accounts for 75% of skin cancer deaths.
• Collected and structured image data from the ISIC (International Skin Imaging Collaboration) dataset, performing train/test splits, and preprocessing with TensorFlow and Keras APIs.
• Engineered a CNN architecture using TensorFlow’s Sequential API, applying data augmentation, and image normalization to enhance model robustness and reduce overfitting.
• Achieved high model performance by experimenting with layers, dropout, and optimization strategies, improving detection accuracy and clinical interpretability.
• Contributed to building AI-driven dermatology support tools, reducing manual diagnostic workload, and enabling faster identification of high-risk patients.

Project 2: Risk Analysis Model - Logistic Regression.| Python | ML Project)

• Performed EDA and reduced multicollinearity via VIF, improving model reliability and stability by ~15%.
• Optimized features using dummy encoding and feature elimination, increasing prediction precision by ~12%.
• Built and Presented logistic regression achieving a ROC-AUC of 0.83, aiding better churn risk classification.
• Enabled customer retention strategies by identifying churn patterns, boosting business engagement outcomes.

Project 3: Intelligent Car Price Estimation Engine - (Linear Regression | Regression | Automotive Analytics)

• Facilitated EDA and outlier detection to enhance model inputs, increasing predictive consistency by ~20%.
• Reduced VIF-related multicollinearity to improve model precision and training convergence.
• Integrated linear regression models with dummy-encoded features, achieving an R² of 0.91 on test data.
• Delivered pricing insights that supported the business in formulating competitive pricing strategies.

Project 4: Cardiovascular Risk Prediction System -(Logistic Regression | Classification | Healthcare Analytics)

• Adapted EDA to reveal health trends, improving feature relevance and downstream model accuracy by ~14%.
• Led multicollinearity via VIF and correlation matrix, enhancing model generalizability.
• Pioneered logistic regression achieving , enabling early detection of high-risk patients.
• Empowered actionable healthcare insights to support preventive interventions and treatment prioritization.