Study description

Heritage Climate Data and Maintenance Priorities: Study Description

About this page. This page describes a study of one heritage building: the question it asks, the data it uses, how “maintenance priority” is defined, and the limits of the study. It does not report how well any model performs, and it does not claim that climate data can predict, or are related to, maintenance priority.

The question

The research asks whether historical climate data can help rank maintenance priorities for a heritage building. In the thesis the aim is stated as developing a predictive maintenance model for cultural heritage sites in Johor Bahru, Malaysia, using machine learning with temperature, relative humidity, precipitation and wind speed as inputs. The thesis lists three research questions: which climate features matter for priority ratings, which machine-learning algorithms suit small datasets, and how the models perform and what rules can be derived from them.

This page describes what the study set out to examine. The thesis words its first question as asking which features “influence” priority ratings; here that is read as something to be examined, and this page does not report an answer.

Source: thesis, sections 1.3.1 and 1.4 (degree thesis, Universiti Teknologi Malaysia).

The building

The Johor Bahru Railway Station Building (Bangunan Stesen Keretapi), Johor Bahru, Malaysia. The published papers describe the work as a single-site case study. The papers’ authors are affiliated with UTM, including the Centre for the Study of Built Environment in the Malay World (KALAM).

Source: thesis section 3.4.2; ICBASE 2025 (Section III); ACM ICBTDS 2025 (Section 4.1).

Climate data: ERA5

• What it is. ERA5 is a reanalysis dataset from the European Centre for Medium-Range Weather Forecasts (ECMWF), accessed through the Copernicus Climate Data Store.
• What was retrieved. Historical values for temperature, relative humidity, wind speed and precipitation, covering 1940 to 2023.
• Where. Values were extracted for the building’s coordinates.
• What it is not. The climate data come from ERA5. They are not readings from sensors installed at the building; the thesis cites the cost of sensors as a constraint of the project. ERA5 is a gridded product, so its values describe a grid cell around a location, not a measurement at the building itself.
• Terminology. The thesis and papers call these variables “microclimate” data. This page calls them climate data, because they are reanalysis values.

Source: thesis sections 3.4.1 and 3.5.1; ICBASE 2025 (Section III); ACM ICBTDS 2025 (Section 4.2). The statement about grid cells is general background on ERA5, not a claim from the thesis.

Building condition data: 38 records

• Where the records come from. Earlier visual condition assessments of heritage buildings in Johor Bahru, taken from student theses held at the UTM Library.
• How many. 38 defect records for the one building, in a table titled as a 2020 defect summary. Each record has a component, a defect, a condition rating, a maintenance priority rating and a “matrix analysis” value.
• What they cover. The assessment scheme lists 16 component types and 37 defect types. The 38 records involve 6 component types (wall 20, column 6, slab 5, ceiling 4, window 2, apron 1) and 12 defect types.
• Rating values in the records. Maintenance priority ratings run from 2 to 5 (2: 5 records, 3: 8, 4: 13, 5: 12). No record has priority 1.
• How the datasets are joined. The thesis describes combining the climate data and the condition records by date and building location. How dates and locations were matched for each of the 38 records is not documented in enough detail to be verified, and this page does not present that matching as audited.

Source: thesis section 3.4.2 and Tables 3.1, 3.2, 3.6, 3.7, 3.8; counts recomputed from Table 3.6 for this draft.

How maintenance priority is defined

• Each component is rated for condition on a 1 to 5 scale (1 negligible to 5 severe).
• It is also given a maintenance priority rating on a 1 to 5 scale, which the thesis describes through the level of consequences of the defect, from no effect on the structure (1) to a defect that may cause structural or service failure if not repaired (5).
• The thesis defines a “matrix analysis” value (up to 25) as condition rating multiplied by priority rating, and groups it into bands from “no maintenance required” to “very severe and require immediate actions”. This is the formula as stated in the thesis. Recomputing from the 38 records in Table 3.6, the value matches the formula in 37 rows; one row (No. 19) does not, and this page does not resolve why.
• These are ratings assigned by inspectors in earlier assessments. They are not physical measurements, and they were not derived from the climate data.

Source: thesis section 3.4.2, Tables 3.3, 3.4, 3.5a, 3.5b.

Approach, in brief

The thesis treats the maintenance priority rating as the target of a machine-learning classification task. The thesis examines five model types: Decision Tree, Random Forest, Recurrent Neural Network, Support Vector Machine and XGBoost. This page does not report or compare their results.

Source: thesis section 3.6.

Limitations of the study

• Small dataset, one building. 38 records from one building. The thesis states that the small dataset limits the robustness of its findings.
• Ratings are inspector-assigned. The ICBASE paper acknowledges that the assessment criteria were not uniformly standardised across the reporting period, which introduces potential subjectivity and variation in ratings.
• Reanalysis rather than on-site data. See “Climate data” above.
• Constraints. The thesis cites the cost of installing and maintaining sensors and limited data availability.
• Not included. The ICBASE paper lists construction typology, material composition and maintenance history (building features) and air quality and pollutant levels (environmental variables) as not included, and notes that the data are historical rather than real-time.
• Directions the authors name. Additional environmental variables (air quality, pollution, soil moisture), a larger dataset built with other organisations, live sensor networks.

Publications

Titles are quoted as published; several contain the word “prediction”. This page does not report their results.

1. Jailani, N. S., Mohamad, M. M., Munajat, M. E., Rahman, N. A. (2025). Maintenance Priority Rating Prediction Based on Microclimate Conditions for Heritage Buildings. ICBASE 2025, IEEE. doi:10.1109/ICBASE66587.2025.11181321
2. Jailani, N. S., Mohamad, M. M., Munajat, M. E., Irawati, I., Nurasa, H., Othman, M. S., Junaidi, A. (2025). Explainable Artificial Intelligence (XAI) for Interpretable Heritage Building Maintenance Prediction. ICBTDS 2025, ACM. doi:10.1145/3786554.3786572
3. Jailani, N. S., Mohamad, M. M. (2025). Machine Learning-Based System for Predicting Maintenance Priorities in Heritage Buildings Considering Microclimate Conditions. Lecture Notes in Networks and Systems. doi:10.1007/978-981-96-5223-5_17 (bibliographic details only; content not reviewed for this page)