Unrealtor

Inspiration

We are a group of wealthy investors looking for the most valuable investments in Toronto real estate. However, our business was hampered by lots of sleazy real estate agents ("realtors") that tried so hard to undermine our profits. So, we turn to technology to get an upper hand in our exploration process.

What it does

Describe a condo unit in our app, tell us the neighbourhood, and we will give you a price prediction.

How we built it

Data Cleaning

Handled missing values, categorical encoding
Mapped each listing to its corresponding neighbourhood
Convert the column from an object data type to a boolean format for improved readability.

Exploratory Data Analysis

Price Trends Across Neighbourhoods 🏡 → Some areas have consistently higher values.
Impact of Property Size 📍 → Larger properties show higher median prices.

Supervised Predictive Modeling (XGBoost)

Why XGBoost? 🔥 Handles large, structured datasets efficiently 🔥 Boosting prevents overfitting 🔥 Works well with categorical & numerical data

Model Training: Feature Selection 📊 → Based on EDA and Correlation Hyperparameter Tuning 🔧 → GridSearchCV Validation Metrics 📈 → R², MAPE

Frontend

We built the app frontend using tkinter and its python binding, making it easier to communicate directly with our backend and made rapid iteration possible.

Challenges we ran into

The limited nature of the data presents a challenge to our team, especially when we want to achieve high prediction accuracy. We also used some unfamiliar technologies in exploratory data analysis such as geopandas and folium, so we used generative AI to help us quickly prototype a working code.

Accomplishments that we're proud of

We made a fairly accurate price predictor resembling the actual prices in the market, with 82.75% of predictions are within a 20% of the actual price. Our model covers a large variance with 0.906 R² score, with good mean absolute percentage error of 11.89%.

What we learned

We are perfecting our skills in doing data analytics, using the skills we learned in class such as EDA, tree classification, and GIS into a small project in just a few hours in a weekend.