Shopping week, though not unique to Harvard, is one of our school’s hallmarks. Allowing students to try before they enroll helps them discover different interests and puts pressure on professors to keep their lectures fresh and exciting.

However, having no hard data about enrollment until more than a week after classes begin is an administrative headache and can make shopping week an ordeal for students as well. Nearly every Harvard student has sat on the floor or stood in the hallway outside of a shopping week lecture because the class proved far too popular for the tiny room it was assigned. The experience is unfair for people shopping multiple classes who can't arrive early enough to get a good seat, and coordinating the eventual switch to a larger room can be difficult or confusing. Worse, a class which is scrambling to find more teaching fellows sometimes has to delay the start of sections or even scale back the number of topics it covers.

Given the complicated mess shopping week has become, it would obviously be helpful to somehow estimate enrollment, even if that means extra work on the part of students. However, the pre-term planning system implemented last fall leaves much to be desired. Many lecture halls overflowed with students during shopping week, and some, such as the popular Statistics 104, were still looking for additional teaching fellows well into the second week.

How did pre-term planning fail so dismally? Some students complained that the planning tool was difficult to use, even with a tutorial video, which may have led to inaccurate data. However, this doesn’t point to a flaw in the concept of pre-term planning, merely the implementation, and given the plethora of design classes Harvard’s Computer Science department offers, our school should certainly be able to create a more user-friendly tool shouldwere pre-term planning to be revisited next year.

A more serious impediment to the accuracy of the data resulted from the pre-term planning deadline being set in November. At that time of year, most students are drowning in a sea of midterms, problem sets and term papers, not to mention extracurricular commitments. Since they would face no repercussions for choosing classes with little forethought or even at random, it is likely that many students did not take the time necessary to carefully consider which classes they would were interested in.

To get more realistic estimates, polling would have to be done after finals, when students finally have a chance to take a break from their work and think further ahead than just the following week. Though January might be on the late side with regard to selecting teaching fellows, it certainly leaves enough time to help inform classroom assignments.

Of course, if data collected in January is too late to be useful, then another option exists for providing estimates from data outside the simple poll. The registrar can instead examine enrollment trends and Q data to more accurately predict how class sizes will change from year to year.

It seems like a no-brainer that a class with no pre-requisites, sky-high Q ratings and a low workload will probably double in size with every passing year. This phenomenon happened in recent years with classes like Digging the Glyphs and Crime and Horror in Victorian Literature. Similarly, a class that had historically low ratings but is being taught by a popular teacher for the first time will almost surely see increased enrollment, which happened with Applied Math 106 last year. A class may also shrink in size when a poorly-reviewed teacher steps in. For example, the optional intermediate physics class in electrodynamics, Physics 153, shriveled down to fewer than ten10 students this semester as even the engineering concentrators required to take it switched to an alternative course offered at MIT.

Of course, cherry-picking specific examples isn't enough evidence to provide the foundation for a highly accurate algorithm, but since the goal of pre-term planning was simply to provide an educated guess about enrollment, finding a system that can generate good estimates shouldn't take an artificial intelligence researcher more than a few weekends worth of work. An undergraduate student could easily do it as a final project for a computer science class, assuming he or she could get access to enrollment data.

Let’s hope that the registrar pursues the data-driven solution for improving pre-term planning, rather than scrapping it, or worse, implementing some sort of pre-registration system. Shopping week may not be perfect, but it’s worth preserving if we can make it better.

Adam R. Gold ‘11, a Crimson editorial writer, is a physics concentrator in Adams House. His column appears on alternate Fridays.