Data set from https://archive.ics.uci.edu/ml/datasets/-Real+estate+valuation+data+set
6 different variables
X1=the transaction date
X2=the house age (unit: year)
X3=the distance to the nearest MRT station (unit: meter)
X4=the number of convenience stores in the living circle on foot (integer)
X5=the geographic coordinate, latitude. (unit: degree)
X6=the geographic coordinate, longitude. (unit: degree)
The output is normalized for the amount of square footage of the house.
The output we’re predicting is (10000 New Taiwan Dollar/Ping (also 3.3 meters)) which is similar to
There was a total of four different algorithms used to try and predict the house’s value. Since we are trying to predict a numerical value, we must use regression model. The first algorithm is a simple linear regression model which preformed with an accuracy of 61%. This wasn’t amazing but it was the first step. The r^2 value was .53 and the mean squared error was 82.1. With 5-fold cross validation the result of the outcome did change significantly. The data then was split from 70% 30% to 80% 20% which did not help the accuracy.
Next was to test the different types of regression models that were available so I used simple vector regression model, linear Bayesian model, and K nearest neighbor.
The best one turned out to nearest neighbor with an accuracy of 67%. This algorithm outperformed all the other ones which relatively low residuals. There are still a few outliers but the general performance is satisfactory.