Chemical Concentration Prediction

Data Science in the Petrochemical Industry

The Problem

• Quality Control Process: Identification of chemical and compound concentration in a product
• Cost of chemical analysis of individual chemical and compounds varies.
• Measurement of some chemicals is expensive

The Objective

• Reduce cost of chemical analysis.

The Solution (Automated AI and Machine Learning)

• Build hundreds of predictive models.
• Individual and ensembles.
• Optimize models.

The Result

An optimized model that predicts concentration of “expensive” chemicals based on the concentration of “cheap” chemicals. Accuracy ~ 94.2%

Prediction of ER Patient Waiting Time

Data Science in the Healthcare Industry

The Problem

• Patients registering at ER desk in a hospital get frustrated from waiting times.
• Frequent questioning of counter staff poses hurdles in ER work.

The Objective

• Improve client experience while waiting.

The Solution

• Process historic data of ER visits, patient history and resource utilization.

• Build hundreds of predictive models.
• Individual and ensembles.
• Optimize models.

The Result

An optimized ensemble of models that predicts waiting time for a patient. Accuracy ~ 86%

Customer Segmentation (for marketing)

Data Science in Marketing

The Problem

• Company has over 10M customers.
• Company has a large bag of products and services.
• Company wants to conduct targeted advertising campaign to sell a specific product.

The Objective

• Identify top 10,000 potential customers.
• Instruct the telemarketing department to promote a chosen product.

The Solution

• Analyze customer usage history.
• Capitalize on demography data and survey results (provided by client).
• Automated Data Mining.
• Identify association patterns.
• Cluster search.

The Result

Provide list of potential customers. Acceptance Rate ~ 72% (reported by client)

Variable Toll

Data Science for Traffic Control

The Problem

• Narrow traffic corridors often experience congestion during peak hours. Especially on toll booths.

The Objective

• Distribute traffic evenly throughout the entire day by linking toll price with predicted traffic.
• Optimize toll booth operations.

The Solution

• Analyze traffic and demographic data of travelers.

• Build hundreds of predictive models.
• Individual and ensembles.
• Optimize models.

The Result

An optimized prediction model to predicts traffic at narrow corridors of a national highway. Theoretical Accuracy ~ 91.2%

Operations Overhead Cost Reduction

Data Science for Operations Monitoring and Cost Reduction

The Problem

• Client has grid of sensing stations over a vast geographic area.
• As the grid grows, so does the operation, monitoring and maintenance costs.

The Objective

• Reduce grid operation cost by eliminating grid points.

The Solution

• Build hundreds of predictive models for each grid point.
• Identify grid points for elimination.
• Replace eliminated grid point readings with predictive model.

The Result

Propose candidate grid points for replacement with prediction models. Cost reduction ~ 37%