Learn from History 101: Simple ways to predict a Data.gov.sg time series using Tensorflow-keras

We’ve all heard the saying: those who cannot learn from history are doomed to repeat it.  But … is it true for a machine?

Time series data is one of the most common types of data we encounter in our daily lives.  Stock prices, temperature measurements, monthly sales numbers are some examples. Data.gov.sg is a repository of Singapore public data sources, and a good local source of time series data.

There are many statistical and deep learning / machine learning techniques in use for time series prediction. Some, like ARIMA, are well-established in statistics and econometrics; others, like Recurrent Neural Networks, are more popular in recent years, but require more work to pre-process and tune.

In this webinar, you will learn a simple but versatile approach of approaching a univariate time series dataset, transforming it to historical features, and training a simple neural network using Tensorflow-keras on Google Colaboratory.  The goal is to give you the basic Lego blocks to perform any time series analysis using Machine Learning.

Note: This is a demo-oriented workshop to teach you the basic concepts. We assume that you have a working knowledge with at least one programming language.