Abstract
The authors describe and explain a family development of the parameter-learning models and algorithms: expert query parametric estimation (EQPE)-based models and checkpoint-oriented algorithms. This class of models and algorithms combines the strength of both qualitative and quantitative methodologies to complement each other to learn optimal decision parameters in an efficient manner to make actionable recommendations. More specifically, this family of models and algorithms relies on domain expertise to select attributes and conditions against the data, from which the family of EQPE-based models and checkpoint-oriented algorithms can learn decision parameters efficiently. To demonstrate the effectiveness and the efficiency of the models and algorithms, the authors have conducted two experimental case studies (i.e., the stock market and the glycemia, respectively). The authors show that their approaches are more effective and produce results that are superior to those of the two other approaches mentioned above.
TopBackground
To support the above decision-making events, there are two approaches, i.e., qualitative and quantitative, that data analysts often use to help domain experts learn optimal decision parameters. A qualitative analysis is a domain-knowledge-based approach that requires domain experts to base upon their past experience and observation to analyze scenarios, from which the experts determine the decision parameters. For example, in the stock market, financial specialists have identified a set of stock indicators that can be used to determine the bear market bottom. The indicators include the S&P 500 percentage decline (SPD), Coppock Guide (CG), Consumer Confidence Index drop (CCD), ISM Manufacturing Survey (ISM), and Negative Leadership Composite “Distribution” (NLCD). If these indicators satisfy the pre-defined, parameterized conditions, e.g., SPD < -20%, CG < 0, etc., it signals that the best period for the investors to buy the stocks, e.g., the S&P 500 Index Fund, is approaching. Often these parameters may reflect some realities since they are set by the domain experts based on their past experiences, observations, intuition, and domain knowledge. However, the suggested parameters may not always be accurate because those parameters are static, but the scenarios that the experts deal with are always dynamic in nature. Thus data analysts need to develop a class of mathematical models and algorithms that can be used to help the domain experts learn the decision parameters dynamically to fit the need of these scenarios.
Key Terms in this Chapter
Checkpoint Algorithm: An algorithm to learn one decision parameter to detect the occurrence of a single event.
Relaxed R-Checkpoint Algorithm: A heuristic algorithm to learn multiple decision parameters, which are fairly close to the optimal parameters learned from the M -Checkpoint algorithm, and to maintain a satisfactorily low time complexity.
Single-Event Expert Query Parametric Estimation Model: A model to learn one decision parameter to detect the occurrence of a single event.
Multi-Event Expert Query Parametric Estimation Model: A model to learn multiple decision parameters simultaneously to detect the occurrence of multiple events in sequence.
Hypoglycemic Expert Query Parametric Estimation Model: A model to learn one blood-glucose-level (BGL) threshold to detect the occurrence of hypoglycemia.
Glycemic Expert Query Parametric Estimation Model: A model to learn two blood-glucose-level (BGL) thresholds to detect the occurrence of both hypo- and hyperglycemia in sequence.
Multidimensional M-Checkpoint Algorithm: A brute-force algorithm to learn multiple decision parameters optimally at their inter-related time points for multi-events.
Linear L-Checkpoint Algorithm: An algorithm to learn one blood-glucose-level (BGL) threshold to detect the occurrence of hypoglycemia.