Predictive analytics is at the heart of data-driven strategies that give organizations their competitive edge. Analytical models for long were based on hindsight. Organizations are moving to advanced analytics and are tightly integrating it into their key decision making process. Predictive Analytics has found many use cases in several industries/functions, to name a few – 1) Marketing to identify customer segments unheard of, 2) Financial Services to prevent fraud, 3) Supply chain to prevent stock-outs during holiday period, 4) Retail to provide seamless & superior omni-channel consumer experience, 5) Oil & Gas to reduce equipment downtime using IoT driven analytics.
At Noah Data, we have a team of highly skilled Data Scientists and Data Analysts with deep domain expertise. They deploy advanced analytics solutions that will help you to conduct pre-mortem on significant investments, reduce fraudulent claims, increase customer acquisition and wallet share, reduce churn, and achieve higher ROI.
Machine Learning / Predictive Models
Survival Analysis | Marketing Mix | Demand Forecasting | Customer Churn Analysis | Segmentation/Clustering Analysis | Recommendation Engines | Pattern recognition | Lead Scoring | Credit Scoring | Outlier Analysis
ARIMA/ARIMAX | Linear Regression / Regression Tress | News Vendor Model | Random Forests | Logistic Regression | Support Vector Machines | k – Nearest Neighbours | Apriori
Agile CRISP-DM (Cross Industry Practice for Data Mining)
Noah Data follows an agile version of the widely practiced CRISP-DM (Cross Industry practice for Data mining) methodology, to develop deployment ready prototypes in shorter sprint cycles.
Interactive workshops are conducted to understand the requirements from a business perspective. The business objectives to be achieved in the current sprint is decided. Based on knowledge gained the data mining problem definition is created.
The initial data and data dictionary (if available) shared are analyzed to get familiar with the data, and to validate the understanding gained from interactive workshops. The Data science team would liaise again with business to deepen their business and data understanding. Data quality issues and first insights are discovered and hypotheses (aligned with the data mining problem defined) are formed based on outcomes of the data analysis.
Prepare data for easy consumption and simplify the process of gleaning insights. This task involves cleansing and transformation of raw data to create the final dataset, and attributes/variables selection, performed multiple times not in any particular order.
Several modelling techniques are chosen to answer the same data mining problem defined. Advanced analytical models, to answer the business question, is developed using each modelling technique, and the identified attributes/variables are calibrated to optimal values. Model specific requirements might warrant stepping back to the data preparation phase.
Accuracy of all modelling techniques is evaluated using appropriate model evaluation error metrics not limited to – Confusion matrix, Gain and Lift charts, AUC – ROC, Gini coefficient, RMSE. The model with the highest accuracy is chosen and run in a small-scale setup to assess the business value created.
If chosen model doesn’t give intended output, then the next CRISP-DM sprint starts from the business understanding phase and continues until a model succeeds evaluation phase.
Deploy the analytical model into operational systems and depending on the data mining problem defined, develop data visualizations to present the insights in clear and concise manner to enable actionable insights by business.
- Deployment ready prototypes in shorter sprint cycles (4-6 weeks)
- Faster time-to-insights
- Iterative discovery and fail-fast approach accelerates analysis
- Closer alignment with business needs