Praveen Raman's Projects
Predicted possibility of customer churning, thereby targeting special campaigns to increase retention rate. Built random forest model to predict churn and developed profit curves to tune parameters. Improved recall rate by 15% from the baseline model through feature reduction, decreasing false predictions.
fast.ai Courses
Improving Dognition's business
Commonly used dot files
Classified traffic signs using Convolutional Neural Networks
Repository for Matlab code Implementation of EKF and UKF for 9-D state space equation
Built machine learning model to predict fraud events based on event details submitted. Compared the performance of Logistic Regression, Random Forest and AdaBoost Classifier Algorithms by fine tuning the model to maximize profit using cost benefit matrix. Developed a web app using Flask, Jinja and CSS to predict the risk level of live data with a recall score of 96% and precision of 85%
Multiclass Problem - Classification of Ghouls and Ghosts
Git Source Code Mirror - This is a publish-only repository and all pull requests are ignored. Please follow Documentation/SubmittingPatches procedure for any of your improvements.
Sequence of code to teach TensorFlow programming
Real Time Twitter Sentiment Analysis using Kafka and Spark
A plugin tracks occurrences and numbers in your notes
Using the user ratings of the Jester Dataset, built a recommendation system with Matrix Factorization. Converted to an Ensemble model, by performing Linear Regression with manually tagged jokes. Error metric decreased significantly. Performed unsupervised KMeans clustering to identify possible groups of joke within data.
Analyzed effect of Emoji's in improving Sentiment Analysis results. Collected twitter data using Twitter StreamAPI and used TF-IDF to vectorize the tweets. Created a positive and negative vector using the matrix, and used cosine similarity to identify the extent to which a given tweet is positive or negative. Incorporated Emoji's to the tweets by converting unicode, and repeated the process. Improved classification of the process by 15%.
Just testing
Used Linear regression model to predict the selling price of a tractor based on parameters which include, age, location and few other parameters. Regularized the model using Lasso method to reduce impact of noise. Primary work was focused on extracting insightful features, which increased accuracy to 70%.