Backend Projects
Overview
This section provides a concise overview of several Python-based projects exploring the power of Artificial Intelligence (AI) across different fields. The first project, focused on machine learning, delves into predicting initial transient voltage within electrical circuits. This application utilizes time-series data and regression models to aid in transient analysis. Shifting to deep learning, another project tackles the diagnosis of diabetic retinopathy in healthcare. Here, convolutional neural networks are leveraged for image analysis to identify the disease. Finally, the data science project investigates explainable AI (XAI) using LIME and SHAP within a healthcare context. This project aims to predict smoking and drinking habits based on body signals. These diverse projects exemplify the practical application of advanced AI techniques and highlight the importance of interpretability, particularly for machine learning models used in healthcare.
My contribution
LIME
SHAP
Fine-tuning models
Exploratory Data Analysis
The team
Alok Karn
Aniket Dixit
Arunbalaji C G
Year
2023-24
Process
Deep Learning Models For Multi-Class Diagnosis of Ocular Diseases Using Fundus Images
Implemented and evaluated CNN models (EfficientNetV2B3, DenseNet201, InceptionV3, ResNet50, Xception)
Enhanced the EfficientNetV2B3 model with multiple dense layers
Added self-attention layers, attaining the highest F1 score of 0.64
Applied models for early ocular disease diagnosis in underserved areas and proposed future improvements for accuracy and reliability
Interpretable Predictive Modeling for Smoking and Drinking Behavior using SHAP and LIME
Evaluated 10+ machine learning algorithms for optimal classification
Applied SMOTE to address the class imbalance and incorporated SHAP/LIME for interpretable AI
Optimized hyperparameters for performance enhancement
Tackled real-world health issues with impactful technology
Analysis of Transient Voltage with Random Forest Regressor
Created a simulated dataset using MATLAB, featuring various network parameters
Predicted the settling time of transient voltage spikes
Conducted a comparative analysis for the prediction of transient voltage and settling time by using 13 regressor machine-learning models
Used time series analysis to further experiment with improving accuracy
