Creative and Business-minded Computer Engineer with 4+ years of experience in developing cutting-edge engineering solutions with a wide range of banking and technology features. Skilled in applying data science to extract valuable insights and engineering useful software features helping the businesses grow. In current role, identified several major bottlenecks and improved software performance by 30% for a Inc. 5000 fastest growing private company. _
This project was done in association with O'Connor & Association. The objective was to build the best model to predict the estimated sale price of commercial properties. Implemntation involved extensive data cleaning and feature engineering followed by modelling. Out of the models tested (Linear Regression, SVR, CNN, Decision Tree, Random Forest, Ensemble Forest, SVR), Random Forest came out to be the best with 24% Coefficient of dispersion compared to 68% of the existing algorithm.
Through this project, I designed a Linear SVM to classify hand language from MNIST data set in GPU and CPU. CUDA implementation of SVM was then compared with CPU implementation using NVIDIA profiler. GPU version took only 22 microseconds comparing to 1298 microseconds of CPU version.
The objective of this project was to detect the pollutant concentration (PM 2.5) and gaseous concentration of SO2, NO, CO2, NO2 for every hour. A geographical model was used to add a feature that described the effect of wind on a location based on the elevation. Although Linear Regression, RNN and Holtz-Winter model was tested, ARIMA model turned out to be best with 94.13% accuracy.
The prediction problem based on a famous Kaggle competition dataset, Bike Sharing Data Set, was implemented using Random Forest in R with a R2-square of 0.8435. The model was fined tuned using grid search to obtain a R2-square value of 0.9586.
