DepNet

Logo
Business Model

Inspiration

Most of current applications of deep learning are in the fields of image recognition, voice recognition, face recognition, or auto-drive. But deep learning is more than that. Many companies are generating more and more data, but how to analyze those data and to obtain insight becomes a large issue. Therefore, we plan to fill this market gap by providing service of data analytics using deep learning with neural networks.

What it does

We train our neural networks in the training dataset and make predictions given inputs.

Due to the limitations of datasets, we obtained Melbourne Housing Dataset from Kaggle.com. And we make the predictions of housing price in specific areas based on the features, such as size of the house, quantity of bedrooms, built year.

With this model, real estate agent can quickly identify the current value of the house and make further decision.

Due the fundamental architecture of neural networks, this model can be easily applied to other markets, such insurance, bank, education, and electricity distribution.

Research themes

Environment, Resources and Sustainability

Creativity, Culture and Society

Data, Knowledge and Decisions

How we built it

Our application is mainly divided into two parts: neural networks modelling and website.

Data Process Since quite a few features in the dataset are categorical values, one hot encoding is implemented to handle this problem.

Neural Networks It is implemented in Jupyter Notebook written by python, with library of keras, pandas, and tensorflow. Due to the limitation of time, we don't design complex modelling such as triplet network, GoogLeNet. We implement multiple layers with Relu activation function.

Since the output is numeric numbers, we implement regression with mean squared error as metrics rather than traditional binary approach with accuracy as metrics.

Website We use website to display our result. When customer input result, the predicted price will be shown. Website example We will deploy website with Neural Networks Modelling on server such as AWS in next step.

Coding Deep Learning and Web Platform

Challenges we ran into

Dataset It took some time to find appropriate dataset. Finally we successfully found Melbourne Housing Price dataset from Kaggle open source dataset.

Server We planned to implement in the AWS server, but due to the limited time we implement our project locally to display.