Inspiration

Buying an HDB resale flat is one of the biggest financial decisions many Singaporeans will make, yet having access to large amounts of transaction data does not necessarily make the decision easier. First-time buyers may still struggle to understand whether a flat they are considering is reasonably priced, whether it is affordable relative to broader household income, and whether there are better alternatives within their budget. We were inspired to build FlatFair to bridge this gap between having housing data available and being able to translate that data into a practical, confident buying decision. This builds on our focus on fair value, affordability and future market conditions rather than historical transactions alone.

What it does

FlatFair is an explainable HDB resale decision-support platform for prospective buyers. A user can enter the characteristics and asking price of a flat they are considering, and FlatFair will estimate its expected fair value and indicate whether the asking price is above or below the model-estimated market value. Rather than showing only a prediction, the platform will also surface comparable historical transactions and the key factors influencing the valuation so that buyers can understand why the model produced that estimate. FlatFair also provides a six-month market outlook for the relevant town and flat type, allowing buyers to understand not only the current estimated value of a flat but also how that market segment may develop. Affordability will be benchmarked by comparing HDB resale prices against Singapore’s median household employment income using a price-to-income ratio. Buyers will also be able to explore alternative town and flat-type combinations that may better match their budget and preferences. This extends the existing concept of personalised valuation, comparable transactions, market forecasts and alternatives into one buyer-focused decision journey.

How we built it

We plan to ingest HDB, data.gov.sg and SingStat datasets into Databricks, store and process the data using Delta Lake, and perform cleaning and feature engineering before creating separate modelling datasets for flat-level valuation and market forecasting. For fair-value estimation, we plan to train a regression model using characteristics such as town, flat type, floor area, storey range, remaining lease and historical transaction information. A separate forecasting model will use monthly town and flat-type trends together with lagged market features to predict resale-price movements over the following six months. MLflow will be used to track experiments, compare models with simple historical baselines and evaluate out-of-sample performance. The final outputs will then be surfaced through a Databricks AI/BI Dashboard and Databricks App, following the architecture already outlined in our proposal.

Challenges we ran into

One of the main challenges we identified is ensuring that FlatFair remains explainable rather than becoming a black-box price predictor. A buyer should not simply be shown a number and be expected to trust it, so we need to present comparable transactions and the main features influencing each valuation clearly. Another challenge is that the datasets operate at different levels of aggregation. Individual resale transactions can be modelled at flat level, while household income is available as a broader benchmark and cannot be interpreted as the actual income of buyers in a particular town. We therefore need to be careful to describe affordability as a comparison against Singapore’s median household employment income rather than as personalised financial advice. Forecasting also introduces a technical challenge because the train-test split must preserve time order and avoid future information leaking into model training.

Accomplishments that we're proud of

At this stage, we are most proud of narrowing FlatFair from a broad housing analytics idea into a focused decision-support product for HDB resale buyers. Rather than trying to serve every possible housing stakeholder, we have centred the concept around a clear user and a clear decision: whether a buyer should consider a particular flat and what alternatives may be available. We have also designed the concept so that its different components work together rather than appearing as separate dashboard features. Fair-value estimation explains the current price, comparable transactions and price drivers provide evidence behind the estimate, the affordability benchmark places the price in a broader financial context, and the six-month forecast gives the buyer forward-looking market information. We believe this creates a more coherent user journey than simply presenting historical charts and predictions.

What we learned

Developing FlatFair has helped us realise that accurately predicting a housing price is only one part of building a useful data product. A technically strong model may still be difficult for a buyer to act on if its output is not understandable or connected to the decision they actually need to make. This led us to place greater emphasis on explainability, affordability context and comparable transactions rather than treating model accuracy as the entire product.

What's next for FlatFair

Beyond the initial prototype, we hope to expand FlatFair into a more comprehensive decision-support tool for HDB resale buyers. Future improvements could include more personalised affordability analysis, recommendations across towns and flat types, and interactive what-if comparisons that help users understand the trade-offs between price, location, flat size and remaining lease.