SCMA 451

Attendance constitutes 5% of the total grade.
Attendance records will be collected at the end of the semester through self-declaration.
- Students will declare what percentage of classes attended and activities participated in.
Full attendance points awarded if 100% attendance claimed and verified against instructor's records.
Purpose: Pedagogical research on the accuracy of self-reported attendance.
- Instructor does not aim to micromanage or distrust students.

Course focuses on applications involving AI, particularly using the R programming language.
Students will function as full stack analysts:
- Given a problem, conduct thorough investigations, process data, create predictive models, and interpret results.

Given with an AI-focused approach.
Midterm exams adjusted to challenge students in an AI-augmented context.
Course design allows students to use AI tools effectively.
Use of AI in real-world coding is emphasized; however, students are encouraged to learn coding independently.

Grade composition:
- Midterms: 55%
- Two midterms.
- Midterm tasks will include practical applications of R in predictive analytics.
- Semester project: 20%
- Assignments: 30%
- Attendance: 5% (suggested as bonus points).
1% bump for student feedback forms filled out by 80% of the class.
- This 1% could significantly impact letter grades.

Students lost points last semester mainly for not following instructions (e.g., not submitting a compiled file).
Students urged to pay attention to detailed work instructions in homework and assignments.
All communications regarding grading issues must be substantiated with argumentation rather than bias.

Students are expected to engage with AI tools in the course.
Different perspectives on the reliability of AI; focus is on how AI can enhance learning.
Emphasis on leveraging competitive AI usage for problem-solving.

Second midterm includes competition question involving training a neural network to predict stock trading decisions.
Evaluation based on model accuracy against randomized holdout datasets.
- Scoring system explained for competition accuracy:
- Top scoring model receives maximum points, subsequent scores adjusted based on distance from top score.

Focus on predictive analytics, regression, and decision trees.
Future topics include:
- Supervised learning (regression, classification) for numerical and categorical predictions.
- Advanced model training and hyperparameter tuning (e.g., hidden layers and their effects on performance).
- Validation strategies to prevent overfitting through cross-validation and robust training methods.
- Exploration of unsupervised learning to uncover hidden data structures.

Majority of workload focuses on data cleaning and processing which typically takes up about 80% of analytics time.
Students will learn to clean datasets, manage missing values, and visualize patterns before modeling.
Data normalization and preprocessing methods will also be covered.

Features and variable selection processes will be taught, including:
- Techniques for scaling (log scaling, standardization via min-max or z-score).
Introduction to regularization methods (e.g., Lasso and Ridge regression) for modeling.
Introduction to decision trees and their nuances including pruning.

The course is designed with respect to students' different majors and their preferences in programming languages (R vs Python).
Suggests exploring models beyond R, including Keras and PyTorch in advanced sessions.
Personal interests mentioned include a background in cooking and sports, creating a relatable and engaging environment for students.

Encouraged students to share names, majors, and interesting personal details to foster community.