ADEC7460 Predictive Analytics / Forecasting , 3 Credits · ADEC7460 Predictive Analytics /...

6
ADEC7460 Predictive Analytics / Forecasting, 3 Credits Woods College of Advancing Studies Spring 2018 Semester, Mar 12-May 15 ONLINE, including synchronous online emeeting Tuesday, 8 – 10 pm ET Instructor Name: Larry Fulton, Ph.D. BC E-mail: [email protected] Phone Number: 210-837-9977 Office: Virtual Office Hours: By Appointment Boston College Mission Statement Strengthened by more than a century and a half of dedication to academic excellence, Boston College commits itself to the highest standards of teaching and research in undergraduate, graduate and professional programs and to the pursuit of a just society through its own accomplishments, the work of its faculty and staff, and the achievements of its graduates. It seeks both to advance its place among the nation's finest universities and to bring to the company of its distinguished peers and to contemporary society the richness of the Catholic intellectual ideal of a mutually illuminating relationship between religious faith and free intellectual inquiry. Boston College draws inspiration for its academic societal mission from its distinctive religious tradition. As a Catholic and Jesuit university, it is rooted in a world view that encounters God in all creation and through all human activity, especially in the search for truth in every discipline, in the desire to learn, and in the call to live justly together. In this spirit, the University regards the contribution of different religious traditions and value systems as essential to the fullness of its intellectual life and to the continuous development of its distinctive intellectual heritage. Course Description This course will expose students to the most popular forecasting techniques used in industry. We will cover time series data manipulation and feature creation, including working with transactional and hierarchical time series data as well as methods of evaluating forecasting models. We will cover basic univariate Smoothing and Decomposition methods of forecasting including Moving Averages, ARIMA, Holt-Winters, Unobserved Components Models and various filtering methods (Hodrick-Prescott, Kalman Filter). Time permitting, we will also extend our models to multivariate modeling options such as Vector Autoregressive Models (VAR). We will also discuss forecasting with hierarchical data and the unique challenges that hierarchical reconciliation creates. The course will use the R programming language though no prior experience with R is required. Textbooks & Readings (Required) Hyndman & Athanasopoulis. Forecasting: principles and practice. Online at https://www.otexts.org/fpp. Available for hard-copy purchase. ISBN-13: 978-0987507105 Coghlan. Little Book of R for Time Series. Available https://a-little-book-of-r-for-time- series.readthedocs.org/en/latest/ R Software (freely available: http://www.r-project.org/ ) Deibold, F. Time Series Econcometrics, a Concise Course . http://www.ssc.upenn.edu/~fdiebold/Teaching706/TimeSeriesEconometrics.pdf Canvas Canvas is the Learning Management System (LMS) at Boston College, designed to help faculty and students share ideas, collaborate on assignments, discuss course readings and materials, submit assignments, and much more - all online. As a Boston College student, you should familiarize yourself with this important tool. For more

Transcript of ADEC7460 Predictive Analytics / Forecasting , 3 Credits · ADEC7460 Predictive Analytics /...

ADEC7460 Predictive Analytics / Forecasting, 3 Credits Woods College of Advancing Studies Spring 2018 Semester, Mar 12-May 15 ONLINE, including synchronous online emeeting Tuesday, 8 – 10 pm ET Instructor Name: Larry Fulton, Ph.D. BC E-mail: [email protected] Phone Number: 210-837-9977 Office: Virtual Office Hours: By Appointment Boston College Mission Statement Strengthened by more than a century and a half of dedication to academic excellence, Boston College commits itself to the highest standards of teaching and research in undergraduate, graduate and professional programs and to the pursuit of a just society through its own accomplishments, the work of its faculty and staff, and the achievements of its graduates. It seeks both to advance its place among the nation's finest universities and to bring to the company of its distinguished peers and to contemporary society the richness of the Catholic intellectual ideal of a mutually illuminating relationship between religious faith and free intellectual inquiry. Boston College draws inspiration for its academic societal mission from its distinctive religious tradition. As a Catholic and Jesuit university, it is rooted in a world view that encounters God in all creation and through all human activity, especially in the search for truth in every discipline, in the desire to learn, and in the call to live justly together. In this spirit, the University regards the contribution of different religious traditions and value systems as essential to the fullness of its intellectual life and to the continuous development of its distinctive intellectual heritage. Course Description This course will expose students to the most popular forecasting techniques used in industry. We will cover time series data manipulation and feature creation, including working with transactional and hierarchical time series data as well as methods of evaluating forecasting models. We will cover basic univariate Smoothing and Decomposition methods of forecasting including Moving Averages, ARIMA, Holt-Winters, Unobserved Components Models and various filtering methods (Hodrick-Prescott, Kalman Filter). Time permitting, we will also extend our models to multivariate modeling options such as Vector Autoregressive Models (VAR). We will also discuss forecasting with hierarchical data and the unique challenges that hierarchical reconciliation creates. The course will use the R programming language though no prior experience with R is required. Textbooks & Readings (Required)

• Hyndman & Athanasopoulis. Forecasting: principles and practice. Online at https://www.otexts.org/fpp. Available for hard-copy purchase. ISBN-13: 978-0987507105

• Coghlan. Little Book of R for Time Series. Available https://a-little-book-of-r-for-time-series.readthedocs.org/en/latest/

• R Software (freely available: http://www.r-project.org/ ) • Deibold, F. Time Series Econcometrics, a Concise Course .

http://www.ssc.upenn.edu/~fdiebold/Teaching706/TimeSeriesEconometrics.pdf Canvas Canvas is the Learning Management System (LMS) at Boston College, designed to help faculty and students share ideas, collaborate on assignments, discuss course readings and materials, submit assignments, and much more - all online. As a Boston College student, you should familiarize yourself with this important tool. For more

information and training resources for using Canvas, click here.

In the case of any technical difficulties or concerns, please contact [email protected] or 617-552-HELP (4357) for immediate assistance.

NOTE: Canvas requires particular computer specifications and wifi access. It is important that you plan accordingly, particularly for courses that have online components.

Course Objectives

1. Students will demonstrate competency across cultural settings and will learn the impact of culture, gender, and age in ADEC 7310 as demonstrated by appropriate synchronous and asynchronous communication. 2. Students will demonstrate ethical competency pertaining to data analysis as demonstrated by assignment submissions and projects. 3. Students will gain intermediate level, practical knowledge of data analysis and econometrics, as demonstrated by assignments and projects. 4. Students will be able to effectively use a statistical/econometric software package, as demonstrated by use of R in assignments and projects.

Grading

• Homework (30%, 3 at 10% each): I expect you to read the material that will be discussed in class (presentation slides, book chapters, notes, manuscripts, etc) prior to the class. I assign homework every week, and that homework covers all material discussed during that week. Homework is due per the calendar (11:59 pm). Homework is individual effort; do not work in teams. I provide solutions to all homework assignments, and grading of homework is on a sliding scale as follows.

0% = No submission 5 or 6% = Below expectation submission, Check Minus 7 or 8% = At expectation submission. Check 9 or 10% = Above expectation submission. Check Plus

• Midterm Project (30%): The midterm project will require you to use material learned in the first part of the course to compete in a data science project against your peers. You will compete in a data science competition and provide a paper / presentation / submissions that measure the accuracy of your forecasts for an international competition. You will provide the following in your paper.

Problem: A discussion of the problem under consideration. Significance: Why is the problem significant or at least interesting? Data: Provide descriptive statistics and graphs. Provide visualization. Discuss processing of the data, cleaning, and feature creation. Literature: FIVE examples from PEER REVIEWED JOURNALS of how the types of models you selected were used in similar situations. Type of Models: What model classes did you build? Why? Formulation: How did you formulate / implement you model in Python / R? Performance / Accuracy: How did your models perform? Limitations: What are the limitations of the model you designed? Future Work: What would you do in the future to improve your models? Learning: What were the key learning points?

You will provide a 5-minute recorded presentation of your model and post this to the assignment forum as well. The rubric follows.

Criteria Ratings Pts

States problem 2.0 pts Good

1.0 pts Poor

0.0 pts Absent

2.0 pts

States significance 2.0 pts

Good 1.0 pts Poor

0.0 pts Absent

2.0 pts

Describes data 2.0 pts Good

1.0 pts Poor

0.0 pts Absent

2.0 pts

Describes types of models 2.0 pts

Good 1.0 pts Poor

0.0 pts Absent

2.0 pts

Reviews literature (5 ref.) 2.0 pts

Good 1.0 pts Poor

0.0 pts Absent

2.0 pts

Formulation of models 2.0 pts

Good 1.0 pts Poor

0.0 pts Absent

2.0 pts

Performance Relative to Peers 10.0 pts

Best 9.0 pts Better

8.0 pts Better

7.0 pts Better

6.0 pts Good

5.0 pts Median

4.0 pts Worse

3.0 pts Worse

2.0 pts Worse

1.0 pts Worst

10.0 pts

Limitations 2.0 pts Good

1.0 pts Poor

0.0 pts Absent

2.0 pts

Future work 2.0 pts Good

1.0 pts Poor

0.0 pts Absent

2.0 pts

What was learned 2.0 pts

Good 1.0 pts Poor

0.0 pts Absent

2.0 pts

Presentation 2.0 pts Good

1.0 pts Poor

0.0 pts Absent

2.0 pts

• Final Project (30%): The final project will require you to use material learned in the second part of the course to compete in a data science project against your peers. The rubric is identical to that for the midterm.

• Discussion Participation (10%): Each student is required to log into the course no fewer than three times each week with no two consecutive days in absentia. Each student is required to post material in the course room on at least two days during the class week, responding to the discussion question by Wednesday at midnight and providing at least two additional substantive posts to colleagues by Saturday at midnight. Comments such as "good post," "nice job," "I agree," etc. are not substantive. The rubric for the discussion is simple. Graduate-level contribution: {0, .5}. Follow-up contributions: {0, .5}.

The graduate grading system for Woods College is as follows: A (4.00), A- (3.67) B+ (3.33), B (3.00) B- (2.67), passing but does not count toward degree C (2.00), passing but not for degree credit F (.00) All students can access final grades through Agora after the grading deadline each semester. Students who complete course evaluations can access grades earlier, as they are posted. Deadlines and Late Work Due to the compressed nature of this course, late homework is accepted up to 5 days after the assignment due date. The penalty is 20% / day. No late examinations or discussion posts are accepted. Course Assignments Most students should spend nine hours each week working to master the content in this course. The weekly schedule and assignments follow. Course Schedule

* T=Forecasting chapter, R=R Textbook chapter, H=homework

Module Topic Readings / Homework

Assigned Homework Due (See CANVAS for specific times and days)

1 Introduction to Time Series & Forecasting, library(forecast) The Forecaster's Toolbox Accuracy Metrics Judgmental Methods Review of Regression Methods

Homework 1 Scan Hyndman, 1-3 Scan R, 1-3 Read Hyndman, 4-5

2 Time Series Decomposition Hyndman, Chapter 6 HW1

3 Exponential Smoothing Hyndman, Chapter 7

4 ARIMA

Homework 2 Hyndman, Chapter 8

MIDTERM PROJECT

5 Manual TS & Forecasting Model Building

Deibold Chapter 5, 6, and 7

6 Advanced Forecasting Methods 1 External regressors, dummy coding, manual model building, Volatility Models

Homework 3 Hyndman, Chapter 9 Diebold, Chapters 8 and 9

HW2

7 Advanced Forecasting Methods 2, VAR models, panel data, Machine Learning

Getting Started in Fixed/Random Effects Models using R

8 Advanced Forecasting Methods 3, Machine Learning & Forecasting Final Project Presentations

All HW3, FINAL PROJECT

Written Work Woods College students are expected to prepare professional, polished written work. Written materials must be typed and submitted in the format required by your instructor. Strive for a thorough yet concise style. Cite literature appropriately, using APA, MLA or CLA style per your instructor’s requirements. Develop your thoughts fully, clearly, logically and specifically. Proofread all materials to ensure the use of proper grammar, punctuation and spelling. For writing support, please contact the Connors Family Learning Center. Attendance Attending class is an important component of learning. Students are expected to attend all class sessions. When circumstances prevent a student from attending class, the student is responsible for contacting the instructor before the class meets. Students who miss class are still expected to complete all assignments and meet all deadlines. Many instructors grade for participation; if you miss class, you cannot make up participation points associated with that class. Makeup work may be assigned at the discretion of the instructor. If circumstances necessitate excessive absence from class, the student should consider withdrawing from the class. Attendance in live sessions is optional but recommended. Consistent with BC’s commitment to creating a learning environment that is respectful of persons of differing backgrounds, we believe that every reasonable effort should be made to allow members of the university community to observe their religious holidays without jeopardizing their academic status. Students are responsible for reviewing course syllabi as soon as possible, and for communicating with the instructor promptly regarding any possible conflicts with observed religious holidays. Students are responsible for completing all class requirements for days missed due to conflicts with religious holidays. Accommodation and Accessibility Boston College is committed to providing accommodations to students, faculty, staff and visitors with disabilities. Specific documentation from the appropriate office is required for students seeking accommodation in Woods College courses. Advanced notice and formal registration with the appropriate office is required to facilitate this process. There are two separate offices at BC that coordinate services for students with disabilities:

● The Connors Family Learning Center (CFLC) coordinates services for students with LD and ADHD. ● The Disabilities Services Office (DSO) coordinates services for all other disabilities.

Find out more about BC’s commitment to accessibility at www.bc.edu/sites/accessibility.

Scholarship and Academic Integrity Students in Woods College courses must produce original work and cite references appropriately. Failure to cite references is plagiarism. Academic dishonesty includes, but is not necessarily limited to, plagiarism, fabrication, facilitating academic dishonesty, cheating on exams or assignments, or submitting the same material or substantially similar material to meet the requirements of more than one course without seeking permission of all instructors concerned. Scholastic misconduct may also involve, but is not necessarily limited to, acts that violate the rights of other students, such as depriving another student of course materials or interfering with another student’s work. Please see the Boston College policy on academic integrity for more information. ©2015 James A. Woods, S.J. College of Advancing Studies