Prasanna Narayana Kone

Recent Data Science graduate with hands-on experience in machine learning, deep learning, and computer vision. Proficient in Python, SQL, PyTorch, and OpenCV, with a solid foundation in statistical modeling and predictive analytics. Developed and evaluated a semantic segmentation model using the Class-Aware Dimension-Pooling Transformer (CADPT) on the ACDC dataset. Additional projects include image classification, macroeconomic forecasting, and insurance claim prediction. Passionate about applying AI techniques to solve real-world problems and contribute to impact, data-driven solutions.

LinkedIn

IEEE Xplore 

GitHub 

Semantic Segmentation Using Class Aware Dimension Pooling Transformer (May, 2025)

• Developed a Class-Aware Dimension-Pooling Transformer (CADPT) model for semantic segmentation on the ACDC dataset under adverse weather conditions (fog, rain, snow, night).
• Integrated Swin Transformer with Dimension Pooling Attention and Class Aware Context Enhancement Module for robust feature extraction.
• Trained and tested the model on the ACDC dataset using PyTorch, achieving a mean Intersection-over-Union (mIoU) of 58%.

Forecasting Macroeconomic Indicators using Time Series Models (May, 2024)

• Built and compared multiple time series models (AR, MA, ARMA, ARIMA, VAR, VARMA) to forecast key economic indicators like RGNP and PGNP, helping analysts interpret long-term trends.
• Identified ARIMA(1,2,2) as the best-performing model based on AIC (927.540), providing the most accurate forecasts for decision-making.