• Aim is to create a data product real-time with low memorywhich is easy to use interface for word prediction similar to SwiftKey.
• Application tries to predict possible 5 best matching words.
• The source data used in this project are files (News, blogs, tweets) taken from the set of corpora provided by HC Corpora [http://www.corpora.heliohost.org]. Complete details bout the data can be found from http://www.corpora.heliohost.org/aboutcorpus.html. Actual source files are available at location <Data set can be obtained from Coursera at https://d396qusza40orc.cloudfront.net/dsscapstone/dataset/Coursera-SwiftKey.zip.
• Only English language files are considered for this project.

• Raw Data is cleaned to remove non alphabetics like numbers, punctuation, etc., converted lower case and sample was created.
• N-grams (1,2,3,4 grams) were created from sample data.
• N-gram rankings/ metrics were used to find the most suitable/ matching words for prediction
• Prediction model is based on the Stupid backoff algorithm (a more simplified approach to Katz Backoff)
• Considers only last 3 words for prediction
• First checks for Quadgram, If not found in Quadgram, it will check for Trigram, like wise it will go to next levels
• Lists 5 best possible words matching

The Shiny applicaitons can be accessed from https://ramch.shinyapps.io/PredictWords/

• Input the word or sentence in the input box
• Application predicts the next potential 5 words on right tab
• Small usage not can be found on left tab below text box
• For performane reasons only last 3 words from the input sentence is considered for predicting the next words
• Pick a word from predicted words or any other word and add it to the input text, on right you will see new set of words.