MartinVVVVVVV/CoCTValuation_2025

This is a dataset used to appeal a valuation completed by the CoCT under SV03 in 2025, for a property in Harfield Village, Cape Town. The city used a CAMA regression to determine that the property value increased from R3.8m to R5.0m, and would not share the parameters of the model. The trigger was the city's sign-off of a renovation and assumes a valuation of R5.0m on 1 July 2022, i.e. the house would have sold for R5.0m in its current state, if it were 2022.

They noted that we must use sales data to argue an incorrect valuation, and provided 2,378 entries within the wider area (Claremont). This dataset includes the above records, additional calculated rows, sub-area segmentation trends and further aggregations to argue a data-driven valuation challenge (a defintion of each field is included below, and the actual challenge submission is included too).

Below is an outline of what we did: - Downlaoded all the Sales Data from the valuation portal (this incldued sales from 2021 to 2025) - All property sale prices we deflated or inflated to 2022 prices (price at 1 July 2022) using the Statistics South Africa Residential Property Price Index (RPPI) for the CoCT. In the absence of RPPI for 2025, deflation was assumed to be CPI (-0.75% per month). - For each, we calculated the price per SQM for an ERF (whole property) and dwelling (SQM of the actual house/dwelling) - The above was plotted on a Google Map using mymaps.google to show the distribution of properies across areas, and this was a key insight, as properties in the sample were spread across multiple areas, including very affluent areas with high rates per SQM - This created the first analysis task. Look at the rates per SQM in our sub-area of claremont (harfield village) vs. rest of the sample, and to do this, we used a nested if statment in Excel to create a marker based on the street name, i.e., if street name = x, mark as HARFIELD - We also looked at how the rate per SQM varied on the size of the dwelling and the ERF, using categories, i.e. 0-100 SQM, 100-200, etc.,m as we wanted to understand if there was variance, and if we could technically rule out a large part of the sample, as our ERF was quite big and generally, larger ERFs have lower rates per SQM - We then looked at

• W

• We plotted every

• Rows: 2,300 (duplicates preserved)

• Source: 23 Sussex Data CoCT SV03.xlsx (converted Nov 2025)

• License: CC-BY-4.0

Load in Code

from datasets import load_dataset
ds = load_dataset("your-username/23-sussex-data-coct-sv03")
print(ds["train"].shape)  # (2300, 13)


![image](https://cdn-uploads.huggingface.co/production/uploads/6918bae6d50641b155fc45a6/lvCJANotE1syWeVnESVLj.png)