Predicting Which Dublin Properties Will Sell

The Problem

Property developers face an expensive guessing game: which homes will actually sell? I took a dataset of 13,320 property listings and built models to predict purchase likelihood, and to identify which factors matter most.

The Approach

1. 1

   Cleaned 13,320 messy records

   Converted text like "2 BED" into numbers, averaged size ranges like "400–500", standardised mixed units into square feet, and filled missing values sensibly (median for numbers, most-common for categories).

2. 2

   Handled extreme values with the IQR method

   Capping unrealistic prices and sizes instead of deleting them, so no data was thrown away.

3. 3

   Trained and tuned four models

   Logistic Regression, Random Forest, KNN, and Decision Tree with GridSearchCV and cross-validation, then compared them honestly on accuracy, precision and recall, not just the headline number.

4. 4

   Found the uncomfortable truth in the results

   All models were good at spotting properties that won't sell, but weak at the rarer "will sell" cases, a class-imbalance problem I documented openly, with fixes proposed (resampling, richer features like proximity to amenities).

The Evidence

The Outcome