Geolocating Alfa Laval's products using supervised machine learning
Bibliographic record
Abstract
A lot of companies have data that can be used to develop a more successful business. To become more data-driven, it is important to extract valuable information from the raw data. One of the largest challenges for companies, while trying to make this transition, is to ensure a data quality at a high level. In this thesis, we worked with Alfa Laval’s database of previously sold products. The main issue with this database was the lack of existing locations, where the products have been installed. In this thesis, we report a solution for the hierarchical prediction of geolocation on three levels: country, city, and coordinates. To build a solution, we examined the three tasks using four different supervised machine learning algorithms. Given our prior knowledge and the available attributes in the database, most tasks proved to yield surprisingly good results. The prediction of countries and cities globally achieved an accuracy of 71% and 57%, respectively. Random forests was the overall best performing algorithm for these two tasks. The prediction of coordinates for the United States was a harder task, resulting in a mean error distance of 872 km, which was achieved by an implementation of artificial neural networks. Our results showed that a prediction of country and city in fact was an achievable goal, even if the existing input did not have an obvious connection to a location. On the other hand, predicting coordinates did not give a result with a sufficiently small margin of error to be useful for most applications.
Fetched live from OpenAlex and de-inverted. Abstracts are not stored in this database: the inverted indexes are 8.6 GB of the frame’s 9.3 GB of text, and the host has 13 GB free.
How this classification was reachedexpand
Full frame machine prediction
Teacher imitationNot calibrated prevalence, not ground truth. Human validation pending. The Gemma side is a direct model label for every work in the frame, read from the title-only record. The Codex side is a classifier learned from the 10,348 direct Codex labels and calibrated to design-weighted sample rates; fields without enough sample support carry no Codex call. Candidate is the union of the two sides; consensus is their intersection. These outputs are machine_predicted_unvalidated and are not human labels.
Distilled classifier scores by category (both heads)
| Category | Codex | Gemma |
|---|---|---|
| Metaresearch | 0.001 | 0.004 |
| Meta-epidemiology (narrow) | 0.001 | 0.000 |
| Meta-epidemiology (broad) | 0.001 | 0.001 |
| Bibliometrics | 0.002 | 0.001 |
| Science and technology studies | 0.001 | 0.000 |
| Scholarly communication | 0.002 | 0.002 |
| Open science | 0.001 | 0.001 |
| Research integrity | 0.001 | 0.001 |
| Insufficient payload (model declined to judge) | 0.003 | 0.005 |
Machine scores (provisional)
The two teacher heads of the student model, read on this work. A score orders the frame for review; it never asserts a category, and the validation status ships verbatim with every row.
Baseline scores from an immature model (maturity gate not passed, 7 training rounds). Scores rank; they never assert a category.
score_only:v0-immature-baseline · verbatim from the scoring run: score_only means the number may rank works, and no category label ships from itClassification
machine, unvalidatedMachine predicted; a candidate call from one source (direct Gemma or distilled Codex), not a consensus.
How this classification was reached, model by model and score by score, is at the end of the page under "How this classification was reached".