OUR END-CLIENT PROJECTS
CASE STUDY 1
Parkinson's UK project
Objectives
To validate the Parkinson's disease phenotype research results by using a range of enhanced data science techniques
To understand how these techniques could be used to further Parkinson’s research with these cohorts
Deliverables achieved
Successfully validated previously identified Parkinson's diseases sub-types
Devised a disease phenotype mapping technique to monitor patient's morbidity status
Click on the video below to see 3D visualization of generated Parkinson's disease classes
CASE STUDY 2
TOUCHNOTE CUSTOMER VALUE PROJECT
Objectives
Understand customer spend and the frequency of purchases
Develop and implement Machine Learning model to predict customer value and purchase frequency
Deliverables Achieved
Developed model for forecasted customer value and forecasted frequency of purchase using a range of variables. These two models are used to categorise customers based on their financial importance
Model predictions were measured as accurate to within £1.30 of actual customer spend
Established an automated weekly process involving model training/learning, validation, prediction and writing the results to a database for business use
Built model front end in the form of RShiny app and Google Sheet to allow business users to review model performance and results
RShiny App developed as pert of Touchnote customer lifetime value project
CASE STUDY 4
CHURN FORECAST PROJECT
Objectives
Understand factors which lead to customer churn (customer stops buying your product/service)
Use Machine Learning to predict the probability of a customer “churning”
Deliverables Achieved
Built a model that predicts the likelihood of a customer leaving based on a range of input variables
Developed an end-to-end automated process to collect data, train and validate the machine learning model on a subset of the data, then apply it to the customer dataset. Finally, write the results to a database
Built a spreadsheet front end so users could review predictions and outputs
CASE STUDY 5
PARKING EYE ROI-BY-SITE PROJECT
Objectives
Identify geo-urban features that correlate with revenue generated from fines
Develop and implement model that forecasts the number of fines issued using a post-code as one of model inputs
Deliverables Achieved
Built a model that predicts number of parking fines based geo-urban features
Used these features to generate a score for each postal area to help Parking Eye identify opportunities for new sites which would yield the best return on investment
Built Rshiny app for business users to review predictions and sort results, and Power BI dashboard to locate and visualise high opportunity areas
POWER BI DASHBOARD FOR CASE STUDY 4: PARKINGEYE PROJECT
Allows users to see the different features and metrics to aid decision making, with colour coded visuals to show relativity between different postal areas for various metrics, and performance of local vs the regional. Aims to facilitate the exploration of various geographical features using a postal area code search. Feel free to play around the dashboard yourself
Designed by Brainbox Data Science LTD & based on non-proprietary public data
CASE STUDY 6
NEWS UK – VARIOUS DATA SCIENCE PROJECTS
Deliverables Achieved
Web Traffic Index full model production – time series analysis to assess and predict web traffic to aid resource management
Article Sentiment analysis – used text analysis program to identify and prove a negative correlation between news article sentiment, and reader engagement
Built a Comment Toxicity model to classify and identify site user comments that breach a threshold of unpleasantness or obscenity
Built a neural network exploring the customer journey from free user to premium subscription
CASE STUDY 7
PWC – DOCUMENT CLASSIFICATION
Objectives
Develop and implement Machine Learning model that classifies Documents
Deliverables Achieved
Developed and presented a range of solutions capable of extracting textual information from documents and predicting document category. For example based on document type, content, author