Please complete the CAPTCHA to continue
@q.com
✓
LinkedIn matched
A concise factual answer block for searchers comparing this professional profile.
John Blankinship is listed as Contract Work: Provide neural network predictive modeling services based in Loveland, Colorado, United States. AeroLeads shows a work email signal at q.com and a matched LinkedIn profile for John Blankinship.
John Blankinship previously worked as Independent contractor/consultant for statistical predictive modeling at Loveland, Colorado and Senior Statistician Manager, Marketing and Analysis Department at Capital One Financial. John Blankinship holds B. S., Physics from Massachusetts Institute Of Technology.
This section adds company-level context without repeating John Blankinship's masked contact details.
AeroLeads found 1 current-domain work email signal for John Blankinship. Compare company email patterns before reaching out.
My name is John Blankinship and I am offering a neural network predictive modeling service. If I cannot build a predictive model that is better than what you already have, there is no charge for my service.I have developed C-language software for the training, validation and delivery of Multilayer Perceptron neural networks. Given a training set of examples, I am confident that I can build a neural network model that will outperform an existing model (neural network or otherwise) with respect to independent validation data — especially for complex data mapping problems with significant nonlinearities or interactions. And I will accomplish this within a week. If the client likes the model's performance, it can be purchased — in the form of a "Weights File" and a C-language function. Otherwise, there is no cost to the client. Key features include:• Neural network models can be applied to classical regression problems, as well as classification and logistic regression problems. For regression problems, neural networks are trained using the Squared Error function with linear outputs. For classification and logistic regression problems, the Cross-Entropy error function is used with logistic or "softmax" outputs.• The Backpropagation algorithm with momentum and online learning is used for training. To facilitate learning down the error surface, the learning rate is dynamically tuned during training.• Using a method called "early stopping," predictive performance is continually monitored during training with respect to an independent "holdout" sample to help avoid overtraining• To help find a superior model and to avoid unsatisfactory local minima, a large number of candidate models are built and tested for 19 different network configurations, 9 different random weight initializations and 3 different learning rates — for a total of 19 X 9 X 3= 513 candidate models. Model selection among the candidate models is based primarily on the predictive performance against the independent holdout sample.• The selected model is validated against both training and validation data (if provided), and includes (1) an analysis of variance, (2) a confusion matrix for classification problems, (3) product moment and rank correlation of predicted vs. observed responses, (4) analysis of residuals, (5) a sensitivity analysis of the importance of each input variable, and (6) scored training and validation data sets• If you are interested, please e-mail me at JB-Nets@msn.com. Sample output deliverables and more detailed information are available upon request.
Listed skills include Neural Networks, Predictive Modeling, Artificial Intelligence, and Statistical Modeling.
A career timeline built from the work history available for this profile.
Loveland, Colorado
As Fraud Consultant for Smith Hanley Consulting, designed potential risk splitters and statistical modeling strategy for client's e-commerce fraud detection system.As independent contractor for client, created SAS programs for promotion response modeling of pharmaceutical sales as a function of six different types of promotion.
Richmond, Virginia
Built and implemented credit risk models for Superprime applications based on credit bureau and income data, increasing approvals by 8% while maintaining same risk level.Developed improved credit risk models for Prime accounts based on credit bureau data and the first 4-8 weeks of transactional behavior.
Richmond, Virginia
Managed group of three statisticians. Responsible for development, implementation and monitoring of predictive models for credit card fraud detection.Built and implemented logistic regression models to score the application fraud risk of new accounts prior to transactional activity, saving $10 million per year in fraud losses. Championed improved strategy for selecting risky accounts based on expected fraud dollar loss rather thanprobability of fraud.Created improved logit models for real-time identity fraud detection based on application-stage data and early transactional behavior, saving $2 million per year.Built and implemented logit model for real-time fraud detection of ATM transactions, saving $1 million per year. Developed logit model for fraud detection of payments based on prior payment and transactional behavior, saving $750 thousand per year.Developed methodology, implemented in SAS, to automate the generation of linear regression models for more accurate budget forecasting of monthly fraud losses.
Based In Greenbrae, California - Worked At Client Sites On East Coast
Provided rule-based expert system development services to mortgage lending industry.For client's online trading system for secondary mortgage market, developed Oracle external stored procedures for price and yield calculations, and rule-based engine for e-mail notification of all trading events. Helped in knowledge acquisition, development and testing of client's rule-based mortgage qualification and pricing system.
Based In Durham, North Carolina - Worked From Home In Virginia
Provided data mining and predictive modeling services to the pharmaceutical industry.Created doctor-level and aggregate promotion response models in SAS to predict a drug's market share as a function of sales force promotion (details and samples).Developed Windows applications in Java for statistical outlier analysis.
Richmond, Virginia
Managed Applied Modeling Group of three statisticians. Responsible for statistical consulting to R&D Dept. and application of statistical methods, neural networks and expert systems to marketing and sensory research, product design and probabilistic risk analysis.Developed direct response (logistic regression) models in SAS to identify prospective and vulnerable consumers for offensive and defensive direct mail marketing.Created C-language and SAS system used for regression modeling, multivariate risk analysis, sensory control regions and trend analysis of all sensory test data.Devised extensive system of C-language software and training procedures for the automated training, validation and delivery of Multilayer Perceptron and Learning Vector Quantization neural networks. Presented methodology at neural network conference.Built neural network models to predict consumer liking as function of brand attributes, product performance as a function of design parameters, brand switching as a function of demographics and product attributes, and pattern classification of chemical data.Implemented expert system for the trend analysis and interpretation of product test data. System was used to help identify probable changes in competitors' brands.
Richmond, Virginia
Built expert system used by machine operators to help diagnose and repair certain manufacturing problems. System was integrated with real-time production data.Designed prototype of expert system for production allocation and planning to help assess the cost, labor and productivity impacts of alternative manufacturing plans.Developed PC-based expert system used for IMS data base design.
Richmond, Virginia
Directed project activities of department of seven analysts. Responsible for development of computer-based decision support systems for operations, financial and marketing management. Provided consulting assistance in areas of production planning, distribution, inventory management, plant scheduling, capital investment analysis and forecasting.Created comprehensive system used for over ten years for the financial analysis of all capital investment projects. Championed adoption of improved profitability measure.Managed development of production planning systems used to generate recommended weekly and long-range manufacturing plans for all brands and plants.Implemented comprehensive linear programming model of company's manufacturing operations to determine least-cost guidelines for long-range production planning.Created simulation model used to determine optimal finished goods inventory levels by brand and warehouse. Formulated improved distribution and inventory policy, saving $1.4 million per year. Directed development of system used to determine recommended daily inventory deployments from three plants to 80 warehouses.
Bethpage, New York
Awarded Engineering Fellowship for two-year work/study program.Helped design space shuttle guidance and control system, and co-authored system description in proposal to NASA. Helped develop software for Air Force to evaluate satellite orbit determination capabilities of ground-based and space-borne sensors.
Quick answers generated from the profile data available on this page.
John Blankinship is listed as Contract Work: Provide neural network predictive modeling services.
AeroLeads has found 1 work email signal at @q.com for John Blankinship.
John Blankinship is based in Loveland, Colorado, United States.
John Blankinship has worked for Loveland, Colorado, Capital One Financial, Mindbox, Inc., Analytika, Division Of Dendrite, and Philip Morris U.S.A.
You can use AeroLeads to view verified contact signals for John Blankinship, including work email, phone, and LinkedIn data when available.
John Blankinship holds B. S., Physics from Massachusetts Institute Of Technology.
John Blankinship is listed with skills including Neural Networks, Predictive Modeling, Artificial Intelligence, and Statistical Modeling.
Search by job title, company, industry, location, and seniority. Export verified B2B contact data when you need it.
Start free trial Search contactsCheck these profiles if this is not the John Blankinship you were looking for.
View similar profiles
