Andres Folleco

Professional Preparation:

• New Jersey Institute of Technology, Newark NJ, Civil Engineering Technology, B.S., 1984
• New Jersey Institute of Technology, Newark NJ, Computer Science, M.S., 1988
• Nova Southeastern University, Davie FL, Computer Science, PhD, 1997

Appointments

• August 2003 – Present, Assistant Professor, Department of Computer Science & Engineering, Florida Atlantic University, Davie, FL
• September 1999 – August 2003, Research Scientist and Visiting Professor, Department of Ocean Engineering, Florida Atlantic University, SEATECH Campus, Dania Beach FL
• December 1993 – September 1999, Technical Staff Member, Sun Microsystems Inc, Plantation, FL

Publications

• Publications most closely related to the proposed project
  1. T. Khoshgoftaar, A. Folleco, J. Van Hulse, “The Impact of Noise on Software Quality Imputation”, (Journal article) in review
  2. T. Khoshgoftaar, A. Folleco, J. Van Hulse, L. Bullard, “Multiple Imputation of Missing Values in Software Measurement Data”, (Journal article) in review.
  3. T. Khoshgoftaar, C. Seiffert, J. Van Hulse, A. Folleco, “Learning with Limited Minority Class Data”, IEEE International Conference on Machine Learning and Applications, Cincinnati, OH, December 2007.
  4. A. Folleco, T. Khoshgoftaar, J. Van Hulse, “Software Quality Classification with Imbalanced and Noisy Data”, Proceedings of the 13th ISSAT International Society of Scientific and Applied Technologies, Seattle, WA, Aug 2007, pp. 191-195.
  5. A. Folleco, T. Khoshgoftaar, J. Van Hulse, C. Seiffert, “Learning from Software Quality Data with Class Imbalance and Noise”, 19th International Conference on Software Engineering and Knowledge Engineering, Boston, MA, July 2007, pp. 487-493.
  6. C. Seiffert, T. Khoshgoftaar, J. Van Hulse, A. Folleco, “An Empirical Study of the Classification Performance of Learners on Imbalance and Noisy Software Quality Data”, IEEE International Conference on Reusability of Information, Las Vegas, NV, Aug 2007, pp. 651-658.
  7. A. Folleco, T. Khoshgoftaar, J. Van Hulse, “Software Fault Imputation in Noisy and Incomplete Measurement Data”, Recent Advances in Reliability and Quality in Design, Springer Series in Reliability Engineering, accepted in April 2007 (Book Chapter).

Professional Interests:

My research interests are within the Data Mining and Machine Learning areas. The areas that I am currently involved include the study of data quality and quality modeling performance and video stabilization and analysis. The research projects I have most recently participated cover the following topics:

• Noise detection, filtering, and correction: The identification of noisy attributes which can easily corrupt and curtail valuable knowledge and information from any given dataset can be critically important to analysts. We have presented several methods and algorithms to detect noisy attribute values and correct them whenever possible.
• Effects of noise and incomplete data on software quality studies: Our studies have quantified the impact of noise on the evaluation of software quality imputation techniques. Several well known imputation techniques were used to impute missing measurement data. A three-way analysis of variance with randomized-complete block design model using the mean absolute error as the response variable was used to analyze the imputation results.
• Classification of ships in surveillance video: This project was part of the Center of Coastline Security Technology federal earmark project. We conducted ship classification as a component of a complete visual surveillance system using data mining and machine learning algorithms and techniques. An empirical study on classifying over 400 instances of ship regions into six types based on their shape features. Shape features were obtained using MPEG-7 region-based shape descriptors.
• Video stabilization from vibration noise: The objective of this empirical study was to provide a software system capable of providing video stabilization from vibration noise with up to three degrees of freedom: translation, rotation, and scaling. A novel periodic correction strategy capable of effectively reducing error accumulation due to vibration noise was produced.
• Learning performance from extremely rare events: Learning from data with rare events is an important, yet infrequently investigated problem. Few if any studies deal with rare events which can account for less than 1% of the training dataset. We examine the performance of several well known classification models when trained on data with rare events. Several variations to the training datasets are also implemented including the number of instances and the use of sampling techniques.
• Learning performance with dependent and independent variable noise: The objective of this empirical study is to quantify the effect of noise in both dependent and independent attributes when learning from such data. The effects of having noisy values in both dependent and independent attributes at the same time will also be studied. This topic will also be explored with several types of datasets with very different statistical characteristics.