DBA 833 Topic 2 data preparation decision log example

Predictive Modeling Grand Canyon University Free custom sample in 24 to 48h

This DBA 833 Topic 2 sample paper logs every change made to a composite freight broker's load records before predicting, at booking, which loads will arrive late. It belongs to GCU DBA 833, Predictive Modeling, listed as DBA-833, a course in the Grand Canyon University DBA. Seven entries cover re-tendered duplicates, time zones, canceled loads, a system migration, filled equipment types with flags, lane codes and post-booking fields that Kaufman and colleagues would call leakage. A reconciliation table traces 412,600 records to 381,400 loads with 57,900 late, about 15.2 percent. Two judgment calls are rerun the other way, and declined loads are named as missing.

CourseDBA 833 Predictive Modeling
TopicTopic 2
Paper typeData preparation decision log
LengthAbout 990 words, 2 pages
FormatAPA 7 student paper
SchoolGrand Canyon University (GCU)
ProgramDBA
UpdatedOctober 2026

Free sample paper for DBA 833 Topic 2

1

From 412,600 Records to 381,400 Loads: A Data Preparation Decision Log for Predicting Late Freight Deliveries at Booking

[Author Name]

College of Doctoral Studies, Grand Canyon University

DBA-833: Predictive Modeling

Topic 2 Data Preparation Decision Log

[Instructor Name]

[Date]

The organization, setting and figures below are a composite written as a model document. No real employer, client, colleague or patient is described.

What this page is doingThe title gives the reconciliation the log documents and the prediction it serves. The broker and every count are a composite, labeled illustrative; the sources are real.
2

From 412,600 Records to 381,400 Loads: A Data Preparation Decision Log for Predicting Late Freight Deliveries at Booking

The Prediction This Preparation Serves

A composite freight broker arranges truckload shipments between shippers and independent carriers. It wants to predict, at the moment a load is booked, whether the load will arrive late, defined as delivery more than one hour after the appointment time, so that its carrier managers can choose a more reliable carrier or warn the shipper. The broker's records hold four years of loads, spanning a change of transportation management system twenty months ago. The moment of prediction limits what preparation may do. Only information available at booking can enter the model, and any field written after booking, such as a carrier's check calls or the actual pickup time, must be removed. Kaufman et al. (2012) called the use of such information leakage and showed how easily it inflates apparent accuracy while producing a model useless in practice. All counts below are illustrative.

Where Preparation Sits

In the CRISP-DM process model, data preparation sits between understanding the data and modeling, and the guide expects it to be revisited as modeling reveals problems (Chapman et al., 2000). Preparation is treated here as part of the model, because each decision changes what the model learns. Every step is logged with its reason, the loads and late arrivals it affected, and whether a reasonable analyst might have chosen differently.

The Decisions, in the Order the Code Runs Them

Entry 1, re-tendered loads. When a carrier drops a load, it is re-tendered to a second carrier and appears twice in the extract, once for each carrier. The two records are merged into one load, keeping the carrier that actually delivered and the original booking time. This removed 18,400 records. Before merging, re-tendered loads were counted as late twice when the second carrier was late, so the late count fell from 61,300 to 58,900.

Entry 2, time zones. Carriers recorded delivery times in their own local zones, while appointment times were stored in the shipper's zone. For loads crossing zones, lateness was misjudged by up to three hours. All times were converted to Coordinated Universal Time before comparison. The conversion changed the late label for 7,850 loads, 4,120 from late to on time and 3,730 from on time to late, a net fall of 390 late arrivals.

Entry 3, canceled loads. Loads canceled after booking have no delivery and therefore no label, and 9,600 were excluded. The exclusion is not neutral: about 2,100 of them were canceled because they were already running behind, and removing them hides some lateness. This is marked as a judgment call.

Entry 4, the system change. For 3,200 loads booked under the old system in its final month, the booking timestamp was overwritten during migration. Without it, the information available at booking cannot be reconstructed, so these loads were excluded, removing 610 late arrivals.

Entry 5, equipment type. Equipment type, such as dry van or refrigerated trailer, was missing for 11,700 loads. It was filled with the type most common on the same lane in the preceding ninety days, and a flag was added so the model can see which values were inferred. Little and Rubin (2019) showed that filling values without recognizing that they were filled can bias later estimates; keeping the flag lets the model treat inferred values differently. This is also a judgment call.

Entry 6, lane codes across the two systems. The old system identified lanes by origin and destination city, the new one by five-digit postal code. Old records were mapped to postal codes using each shipper's facility address, which matched 97 percent of loads; the remaining loads kept a city-level lane and a flag marking the coarser code. Without this step, the model would have treated the same lane before and after the migration as two different lanes, splitting its history in half and making older loads look less informative than they are.

Entry 7, post-booking fields. Eleven fields written after booking, including check calls, actual pickup time and detention charges, were dropped from the modeling file. No loads were removed.

Reconciliation

The load count can be traced from the raw extract to the modeling file without gaps. The prepared file holds 381,400 loads, of which 57,900, about 15.2 percent, arrived late.

Table 1

Load and Late Arrival Counts From Raw Extract to Modeling File, Illustrative

StepRecords or loadsLate arrivalsChange
Raw extract412,600 records61,300
1. Merge re-tendered loads394,20058,90018,400 records merged
2. Convert time zones394,20058,5107,850 labels changed, net 390 fewer late
3. Exclude canceled loads384,60058,5109,600 removed; about 2,100 were running late
4. Exclude loads with lost booking time381,40057,9003,200 removed
5. Fill equipment type, add flag381,40057,90011,700 values filled
6. Map lane codes across systems381,40057,90097 percent matched to postal codes
7. Drop post-booking fields381,40057,90011 fields removed

Two Judgment Calls Rerun the Other Way

Counting the 2,100 canceled loads that were running late as late arrivals, instead of excluding them, would raise the late share to about 15.6 percent and shift some weight toward the lanes and carriers where cancellations cluster. Leaving missing equipment type as its own category instead of filling it changed the validation ranking quality of a first model only slightly, by less than half of one percent. The first choice matters more, and the modeling report will present results both ways.

What the Prepared File Cannot Represent

The file describes only loads the broker booked. Loads it declined, because no carrier would take them at the shipper's rate or the lane was known to be unreliable, never appear, and they may be among the likeliest to run late. The model will learn lateness among accepted loads and should not be used to decide which loads to accept. The file also stops at the system change for one month of bookings, so any seasonal pattern in that month is slightly underrepresented.

References

Chapman, P., Clinton, J., Kerber, R., Khabaza, T., Reinartz, T., Shearer, C., & Wirth, R. (2000). CRISP-DM 1.0: Step-by-step data mining guide. SPSS.

Kaufman, S., Rosset, S., Perlich, C., & Stitelman, O. (2012). Leakage in data mining: Formulation, detection, and avoidance. ACM Transactions on Knowledge Discovery from Data, 6(4), Article 15. https://doi.org/10.1145/2382577.2382579

Little, R. J. A., & Rubin, D. B. (2019). Statistical analysis with missing data (3rd ed.). Wiley.

DBA 833 Topic 2 instructions, in plain terms

The Topic 2 assignment in DBA-833 asks doctoral students to document data preparation as part of the model. Strong logs begin from the prediction and the moment it is made, since that decides which records and fields may be used. Students should record each decision with its reason and the number of records and positive cases it changed, mark the choices a careful analyst could make differently and rerun some of them the other way. The log should reconcile counts from the raw extract to the modeling file and state what the prepared data can no longer represent.

How this DBA 833 Topic 2 example is built

The finished log treats preparation as part of the model rather than a chore before it. The broker wants to predict, at the moment a load is booked, whether it will arrive late, and its records hold several years of loads across a change of transportation management system. Each decision gets an entry. Loads re-tendered to a second carrier appear twice and are merged into one. Delivery times recorded in each carrier's local zone are converted to a single standard. Canceled loads are excluded, and the log states that this also removes loads canceled because they were already running behind. Missing equipment types are filled from each lane's usual pattern, with a flag kept so the model can see which values were inferred. Each entry reports how many loads and how many late arrivals the step removed or changed.

Reading the DBA 833 Topic 2 rubric

Faculty reward preparation treated as modeling. An Excellent log places each step in CRISP-DM and removes post-booking fields as leakage, following Kaufman and colleagues. Credit goes to entries giving loads and late arrivals changed for duplicates, time zones, cancellations and the system migration. The rubric values filled equipment types kept with a flag, as Little and Rubin advise, and a reconciliation table from 412,600 records to 381,400 loads. Strong logs rerun two judgment calls. Top marks go to naming declined loads as data the file cannot hold.

DBA 833 Topic 2 help: mistakes that cost marks

The largest loss here is a preparation section that says the data was cleaned and gives no entry for what cleaning meant. A reader cannot reproduce a result built on choices nobody recorded, and at the doctoral level an unreproducible result carries little weight. Exclusions described without their effect on the outcome rate are the next problem, because dropping canceled loads can quietly lower the share of late arrivals the model learns from. Imputation performed silently lets filled values pass as observations. Statistics for imputation or scaling computed on the whole file borrow information from the period later used for testing. Logs that present every choice as obvious hide the judgment calls a reviewer would most want to test, and a log with no load count reconciled from start to finish cannot show that nothing went missing.

Get a DBA-833 Topic 2 example written to your instructions

Send the DBA-833 Topic 2 instructions, your classroom rubric and a description of the dataset your section is preparing. A custom example is written to those criteria, with the prediction moment stated, every decision logged with its effect, inferred values flagged, judgment calls rerun and counts reconciled, returned in 24 to 48 hours. The first one is free.

More DBA 833 papers and DBA samples

DBA 833 Topic 2 questions, answered

Why does data preparation take so much of the effort?

Because operational records are collected to run a business, not to train a model. Shipments are re-tendered, systems change their field definitions, times are logged in different zones and some values are never entered. Each problem needs a decision, and each decision changes what the model will learn. Practitioners commonly describe preparation as the largest share of a project, which is why the course gives it a topic of its own.

Should preparation steps be fitted on all of the data?

No. Values used to fill gaps or rescale variables should be computed from the training data only and then applied to later records, because computing them across every record lets information from the test period shape the training file. Performance then looks better than the model will manage on new loads. The example computes each such value inside the training period and records that it did so.

What counts as a judgment call in preparation?

A choice on which careful analysts could reasonably differ, such as whether a delivery rescheduled by the receiver counts as late or whether an unusually long transit time is an entry error. The example marks those choices, reruns two of them the other way and reports whether results change. Showing that a conclusion survives a reversed choice is stronger evidence than asserting that the chosen option was correct.

Where can I find a free DBA 833 Topic 2 sample paper?

The decision log is printed here in full: the prediction it serves, where preparation sits, seven entries, the reconciliation, two judgment calls rerun and what the file cannot represent. Your own DBA 833 Topic 2 prompt gets a free log.

Why keep a flag on filled values in DBA 833 Topic 2?

Filling missing equipment types from each lane's usual pattern makes inferred values look like recorded ones. The flag lets the model treat them differently, which limits the bias that unacknowledged filling can create.