In this final part, I will share some popular machine learning algorithms to predict the housing prices and the live model that I have built. My objective is to find a model that can generate a high accuracy of the housing prices, based on the available dataset, such that given a new property and with the required information, we will know whether the property is over or under-valued.
Brief introduction of the machine learning algorithms used
I explore 5 machine learning algorithms that are used to predict the housing prices in Singapore, namely multi-linear regression, lasso, ridge, decision tree and neural network.
Multi-linear regression model, as its name suggest, is a widely used model that assumes linearity between the independent variables and dependent variable (price). This will be my baseline model for comparison.
Lasso and ridge are models to reduce model complexity and overfitting when there are too many parameters. For example, the lasso model will effectively shrink some of the variables, such that it only takes into account some of the important factors. While there are only 17 variables, in the dataset and the number of variables may not be considered extensive, it will still be a good exercise to analyse the effectiveness of these models.
Decision tree is an easily understandable model which uses a set of binary rules to achieve the target value. This is extremely useful for decision making as a tree diagram can be plotted to aid in understanding the importance of each variable (the higher the variable in the tree, the more important the variable).
Last, I have also explored a simple multi-layer perceptron neural network model. Simply put, the data inputs is put through a few layers of “filters” (feed forward hidden layers) and the model learns how to minimise the loss function by changing the values in the “filters” matrices.
#predictive-analytics #predictive-modeling #machine-learning #sklearn #housing-prices
