1. Use Case - ML Analytics Solution for financial services customer – Credit Score & Credit Limit
Client Challenge:
- To improve customer base and product over new trends and investment possibilities.
Our Solution and Approach
Credit scoring systems; This trend is based on the volume of information collected and used as predictors in Machine Learning Models. Social media interactions and other data was used as additional sources of information in risk management.
- Understand current state of Credit Limit system
- Define MVP scope
- Defined Road map to desired state
- Data Providers evaluation and assessment
- Design Options for experimentation platform for the client
- Define solution architecture
- Train Models
- Choose optimal algorithm and methodologies
- Agree and deliver MVP, Seek Feedback and Refine
- Confirm Data partnerships to be leveraged
- Build Experimental platform
Benefits we delivered to client
- Define solution architecture
Nandini Consulting Lessons Learnt
- Integration with existing systems must be treated with sensitivity
- The human element — be sensitive and manage expectations
- Sourcing right datasets from various data sources
2. Use Case - Sentiment Analysis Analytic Solution for Telecom Customer
Client Challenge:
- Organization’s poor quality of data and inaccessible as it was stored in siloes on multiple legacy systems.
- Large data sets and integration of data sources on cloud platform
Our Solution and Approach
We used a Deep Learning technique called Recurrent Neural Networks (RNN); implementing it using Long Short Term Memory (LSTM) Network Architecture.
Step-by-step process:
- Load in and visualize the data
- Data Processing — convert to lower case
- Data Processing — Remove punctuation
- Data Processing — Create list of reviews
- Tokenize — Create Vocab to Int mapping dictionary
- Tokenize — Encode the words
- Tokenize — Encode the labels
- Analyze Reviews Length
- Removing Outliers — Getting rid of extremely long or short reviews
- Padding / Truncating the remaining data
- Training, Validation, Test Dataset Split
- Dataloaders and Batching
- Define the Long Short Term Memory (LSTM) Network Architecture
- Define the Model Class
- Training the Network
- Testing (on Test data and User- generated data)
Benefits we delivered to client
- Better understanding of customer behavior
- Better product and service offerings
- Customer retention
Nandini Consulting Lessons Learnt
- Language complexity, and a process to quantifying and scoring language
- Analyze competition sentiment, in addition to your own
