Applied Machine Learning / Data Science

Applied Machine Learning / Data Science

With the advent of pathbreaking technologies such as artificial intelligence, machine learning, big data analytics, blockchain / distributed ledger technologies and innovation platforms, we have recently begun to also expand my research scope to better understand how these novel technologies and organizational adoption of related business practices will impact organizational capabilities for value creation.  The research here follows a design science approach along with empirical and computational validation.  

E-Commerce Analytics

The focus here was to develop an approach for quantifying the impact of systems design on how information systems are actually used.  We empirically evaluated the impact of e-commerce website design on consumer purchasing behaviors, developed conceptual frameworks for linking systems design to measurable business value implications by developing models of online consumer behavior in the context of Internet-based selling websites, and data-driven analytical methods for assessing the business value impacts of e-commerce website design.  

Real-Time Bidding in Mobile Digital Advertising

In mobile digital advertising, advertisements are served in real-time through an auction mechanism.  In order to improve the return on investment (ROI) of mobile advertising campaigns, demand-side platforms must determine which ads to bid for that maximizes apposite reach and engagement likelihood (e.g., clickthrough).  However, this is a non-trivial problem as the mobile advertising ecosystem exhibits extreme dynamism that prevents standard machine learning models to be readily applied.  We developed a two-phased approach and algorithm composed of intelligent bid selection and supervised bid selection that maximizes the balance between reach and engagement in the face of dynamic features and feature values.  

Handling Missing Values in Data Analytics

One of the most enduring problems in data analytics is the problem of missing data.  Even with the advent of the era of big data, the missing data problem does not lessen.  Most researchers and practitioners implicitly believe that because they have an abundance of data, they can simply discard observations with missing values and assume that the conclusions drawn from the data analysis will be unbiased and efficient.  However, although this may be the case if the mechanism of missingness is missing completely at random (MCAR), this is not the case, if the missingness mechanism is missing at random (MAR) or not missing at random (NMAR).  We propose and develop an estimation approach that explicitly incorporates the missingness model into the likelihood function and use Monte Carlo-based computational methods to parameter estimation.  

Algorithmic Decision-Making and Governance

Traditionally, the evaluation of information systems design could be conducted at the design stage, with design blueprints even before the information system is developed and deployed.  However, with the advent and wide-spread deployment of artificial intelligence (AI) and machine learning (ML)-based technologies, where algorithm-based automated decision-making are taking over human decision-making for many important tasks (e.g., product recommendations, hiring decisions, self-driving cars, validation of financial transactions, etc.), a situation referred to as algorithmic decision-making and governance, this has become more difficult, if not impossible.  Therefore, it is important to develop methods to a-priori evaluate the performance of such information systems, rather than post-doc detect problems in run-time while the information system is in production mode.  We have developed an agent-based simulation model to study algorithmic governance in consensus mechanisms in blockchain technologies, which allows us to develop design theories for algorithmic governance.