Kunal Krishna

Provided a service to register Raspberry Pi device and can control led light. Used Mosquitto message Broker, MQTT protocol, MySQL, Flask, Python, AWS Ubuntu EC2 instance to host a webserver, Log visualization.

Goal: Show real-time sentiment analysis based on Geographical Location

Technology Stack: Spark Streaming, Spark ML-lib, Kafka, Java, Scala, NTLK, python, ElasticSearch, Kibana, Natural Language Processing and Machine Learning

Learning: Real-time data processing using Kafka by creating topic based on a tweet and spark streaming subscribe to that topic to process data.
Applied Natural Language Processing and machine learning (Spark- MLlib) to get sentiment
Create Index using… Show more

Goal: Show real-time sentiment analysis based on Geographical Location

Technology Stack: Spark Streaming, Spark ML-lib, Kafka, Java, Scala, NTLK, python, ElasticSearch, Kibana, Natural Language Processing and Machine Learning

Learning: Real-time data processing using Kafka by creating topic based on a tweet and spark streaming subscribe to that topic to process data.
Applied Natural Language Processing and machine learning (Spark- MLlib) to get sentiment
Create Index using geo-location and sentiment for each tweet using ElasticSearch
Used Kibana as a visualization tool. Show less

A real time business requirement from a real time use case was taken as the subject of this project. The aim of this project to find out which cloud provider should we choose, how many nodes we required, virtual machine instance configuration for a different number of the concurrent user. We have to use Cassandra NoSQL database. Finally, based on performance and cost (created a graph) provided details document of YCSB benchmarking and solved all the requirement.

The learning inculcated… Show more

A real time business requirement from a real time use case was taken as the subject of this project. The aim of this project to find out which cloud provider should we choose, how many nodes we required, virtual machine instance configuration for a different number of the concurrent user. We have to use Cassandra NoSQL database. Finally, based on performance and cost (created a graph) provided details document of YCSB benchmarking and solved all the requirement.

The learning inculcated through this project:
1) Learn how to launch virtual machine instances on each IaaS providers.
2) Learn how to install/use NoSQL storage for Big Data, like Cassandra.
3) Learn how to install YCSB benchmark, compare performance, and analyze it to support business decision.
4) Compare prices and features provided by each IaaS providers, and decide who is better for a given scenario.
5) Created a visual comparison document in the form of performance and price graphs. Show less

Aim: Predict digit from handwritten digit image
Used Simple computer vision dataset(mnist), generated multiple hidden layers to create a computational graph. This computational graph act as input and then used Tensorflow library to get the accuracy.
Process data in batch and each batch size is 100.
Achieved 95.52 % Accuracy.

Technology: TensorFlow, Python, DeepLearning

Hosted a personal website, implemented auto scaling for client server program, created S3 bucket to store images, IAM to create user group. Used Amazon EC2, IAM, S3, Cloud Watch to complete this project.

Implemented Map-Reduce facility, with master and worker nodes for reduce-side join, map-side join, distributed caching, job chaining technique over large Yelp Dataset. Also used Hive to answer the same query to test the optimization and run-time difference.

Implemented Naive Bayes algorithm to predict the party information of a candidate like if they belong to Republic or Democratic based on the option they have selected on the different issue like social, health, education. Used R programming and Rattle tool to do more analysis on attributes and their relation using PCA and co-variance.

Implemented Natural Languages Processing algorithms of finding the correct tag for a word, its probability of occurrence
and also the probability of a sentence of certain keywords occurring together

Used KNN algorithm to build recommendation system using Netflix data-set in java. below approach has been used:

The ITEM_BASELINE Model: guess that a new rating for an item will be about equal to the average rating that that item has received in the past.

The RATER_BASELINE Model: guess that a new rating for an item will be about equal to the average rating that the user you're looking at has given in the past.

The MIXED_BASELINE model: make a good guess about what rating… Show more

Used KNN algorithm to build recommendation system using Netflix data-set in java. below approach has been used:

The ITEM_BASELINE Model: guess that a new rating for an item will be about equal to the average rating that that item has received in the past.

The RATER_BASELINE Model: guess that a new rating for an item will be about equal to the average rating that the user you're looking at has given in the past.

The MIXED_BASELINE model: make a good guess about what rating you'd expect the item to receive and what rating you'd expect the rater to give, and split the difference between them.

For the cross-validation parameters, I kept it at 10 cross folds and used the first fold as my test set (this leads to a test set of size 10.000).

To calculating similarities,Applied Euclidean distance and Pearson co-efficient to find the nearest neighbor. For my custom method, I take a weighted average between the item-based score
and the user-based score.

Finally as per my experiment, the rater-based prediction is better than the item-based one - at
least for these parameters.

Language : Java Show less

Designed a Database host application that interfaces with a backend MySQL database implementing an Online Library Management System.

Implemented following functional requirements in my web application:

a) Book search and availability

b) Checking out books

c) Checking in books

d) Borrower Management (Add borrower)

e) Calculate fines

Project Environment: JSP, Servlets, HTML5, MySQL Database, MySQL Workbench, Eclipse IDE for JAVA EE

An algorithm that can flag emails as spam using Rtexttools. Implemented Naive Bayes, Logistic Regression,SVM, Decision Tree and Perceptron algorithms in R and compared performances of the algorithms.

The project was based on information retrieval.

Crawled 100000 pages and used indexing, compression and query expansion to display results for search queries on books and compared them to results of Google and Bing with respect to performance and relevance.

Created Inverted Index for the search engine, used JUNG framework to create web graph. Created two Relevance model using vector space and based on Page Ranking, HITS which helped us to do link analysis. Used Gamma, Delta… Show more

The project was based on information retrieval.

Crawled 100000 pages and used indexing, compression and query expansion to display results for search queries on books and compared them to results of Google and Bing with respect to performance and relevance.

Created Inverted Index for the search engine, used JUNG framework to create web graph. Created two Relevance model using vector space and based on Page Ranking, HITS which helped us to do link analysis. Used Gamma, Delta encoding system to compress the index. Show less

Predict user Behavior based on his/her purchase history. We used data-set which has information related to all the user who has used promotional coupon and then come back to do shopping again on the same website. As part of data pre-processing we handled null value, junk value,missing value.

Fill in missing values (attribute or class value):
Ignore the tuple: usually done when class label is missing.
Use the attribute mean (or majority nominal value) to fill in the missing… Show more

Predict user Behavior based on his/her purchase history. We used data-set which has information related to all the user who has used promotional coupon and then come back to do shopping again on the same website. As part of data pre-processing we handled null value, junk value,missing value.

Fill in missing values (attribute or class value):
Ignore the tuple: usually done when class label is missing.
Use the attribute mean (or majority nominal value) to fill in the missing value.
Use the attribute mean (or majority nominal value) for all samples belonging to the same class.
Predict the missing value by using a learning algorithm: consider the attribute with the missing value as a dependent (class) variable and run a learning algorithm (usually Bayes or decision tree) to predict the missing value

Normalization:
Scaling attribute values to fall within a specified range.

Data reduction
Reducing the number of attributes
Principle component analysis (numeric attributes only): searching for a lower dimensional space that can best represent the data

Reducing the number of attribute values
Binning (histograms): reducing the number of attributes by grouping them into intervals (bins).
Clustering: grouping values in clusters.
Aggregation or generalization

Used Bragging and boosting approach to increase the accuracy.

Used Rattle, R programming language to perform Statistical modeling which helped to predict user behavior.

Received professor appreciation for using a different technique to predict the behavior. Show less

Idea is similar to an old arcade game called Breakout, with some variation. Used concept of multi- threading and android motion sensor , accelerometer. In order to complete this game need to break three layer of bricks and touch the top part of the screen.After completing the game user will have option to store their highest score.

Technology and Tools: Android SDK, Java,XML,Android programming concepts.

Created a database using one million records of Majestic Million CSV file. Main Idea of this project is to implement Database indexing and web programming. Developed a web interface were we provided option to query a domain name, then get its ranking, also query domains belong with
certain TLD. Created a different table based on the unique index and different web page for the same. Used CURL utility to test the website performance and also Apache/JMeter to check how many concurrent user our… Show more

Created a database using one million records of Majestic Million CSV file. Main Idea of this project is to implement Database indexing and web programming. Developed a web interface were we provided option to query a domain name, then get its ranking, also query domains belong with
certain TLD. Created a different table based on the unique index and different web page for the same. Used CURL utility to test the website performance and also Apache/JMeter to check how many concurrent user our website can handle.

Technology and Tools: Java, Node.js, HTML, CSS, Oracle, SQL, Sell scripting Show less

Contact Manager application similar to our android contact manager. Option to add, update,delete contact information of user. Used text file to store the back-end information. Idea of this project to give more emphasis on User Interface design principle, hence followed the android design principle.

Technology and Tools: Java, Android SDK, XML, Android Concepts

Aim: Analyze the sentiment for a tweet
BaseLine: Created Bag of words, and calculated the probability based on the occurrence of words in tweets to determine a sentiment.

Bag of words was determined as follows:
1) We took synonyms of four types of emotions (angry, sad, happy, fear) we were analyzing a tweet on.
2) To increase the size of the bag we downloaded tweets based on the words we got in the first step. Thereafter we applied lexical feature like tokenization, stop word… Show more

Aim: Analyze the sentiment for a tweet
BaseLine: Created Bag of words, and calculated the probability based on the occurrence of words in tweets to determine a sentiment.

Bag of words was determined as follows:
1) We took synonyms of four types of emotions (angry, sad, happy, fear) we were analyzing a tweet on.
2) To increase the size of the bag we downloaded tweets based on the words we got in the first step. Thereafter we applied lexical feature like tokenization, stop word removal, Lesk and synonym features to increase the bag of words by 50%.

Improvement Strategies on BaseLine:
1) Lexical: Applied tokenization, stop word removal, used regular expression to remove URL link, created abbreviation file to replace internet jargon to that word, for example: lyk:Like or lv:love. Removed repeated character , for example: havvve to have. Improved accuracy: 10%
2) Syntactic: Applied chunking and part of speech tagging selected verb, adjective, noun and adverb. Improved accuracy: 25%
3) Semantic: Applied Lesk and synonymy feature: Improved accuracy: 25%

After Applying all the NLP feature one by one we got a feature vector. To test this feature vector we created learner model using Navie Bayes algorithm, for this we have manually filtered 200 plus tweets related to each sentiment and train this model. Then we used our feature vector and model to determine sentiment of tweet.Improved accuracy: 20%

Language and Tool: Python and NLTK, Pycharm Show less

Developed a website for Online Shopping using HTML,CSS,PHP and MySQL on Dreamweaver and deployed it on WAMP server. The website has provision for online shopping application, share product review to other user and other basic function.

Provided a platform for job seekers and job providers wherein they can search and post the job. It also included many other basic functionalities using ASP.NET.