Neha Pawar

Built a framework using Apache Spark to preprocess 450GB data/day and created 4 forecasting models for telecom tower cells’ Throughput and Latency prediction using LightGBM and Catboost Regressor which achieved precision > 70%. Designed and implemented an energy consumption model using Linear Regression that helped in achieving higher energy saving up to 10-15% and improved customer engagement on the platform. Developed interactive graphs on a web app to enable the end-user to visualize and analyze the data. Optimized the cell performance 2-2.5 times by predicting PRB Utilization using Linear Regression which uses KPIs and Google Map data. Built 5 multi class text classification model using weighted TF-IDF Vectorizer and Neural Network which achieved accuracy of approx 80% and deployed it in production that helped to increase the revenue by 10% and reduced the manual time by 50%.

Developed a text extraction model for a particular domain using DBpedia Bangalore, Karnataka, and other resources and implemented domain-specific Name Entity Recognition framework.

Built a defaulter merchant prediction classification model using Logistic Regression which achieved an accuracy of approx 82%. Scaled the product to enable it to run efficiently on the distributed environment using Apache Spark. Built the RESTful services to enable the clients to perform the background verification of their end-users.

Created the initial version of the recommendation system that was integrated into the product for employers to identify potential candidates for a particular job description. Developed a recommendation system for the applicant based on various candidate parameters such as previous work experience, skills set, etc.

Thesis title: Social Network based Recommender System. Detecting overlapping communities formed by users in the social network using the Network Decomposition method. Once detected use this community information for recommending an item to the target user.