As the name suggests, an ordered set of observations made over a period of time is time series. Since time-series contain sequential data points mapped at successive time duration, it can be a very important tool for making predictions. Some of its major application areas include — stocks and financial trading, analysing online and offline retail sales, and medical records such as heart rate, EKG, MRI, and ECG. Time series data is one of the most common types of data that is available today. These data can vary between a person’s annual salary fluctuations to stock market values.
Below we list five open-source machine learning time series projects, in no particular order, for enthusiasts to try their hands on:
Time Series Analysis of Inflation Rate Using Shinyboard
The goal of this project is to analyse and study the inflation rates of countries and major economic unions around the world. The dataset used for this project can be taken from the public data made available by the International Monetary Fund (IMF). In this project, particularly, data from 1980 to 2017 was, and the projected inflation rate of countries till 2022 was obtained. The shiny dashboard was made in R.
Visit here for more information.
Internet Traffic Forecasting Using Time Series Methods
With this project, the amount of traffic on TCP/IP networks can be forecasted using time series forecasting method. The data was collected from two internet source providers and was analysed using different ahead predictions and time scales. The project used two time series methods — ARIMA and Holt-Winters. In the original experiment, the performance of traffic prediction with time series method was compared against a novel neural network ensemble approach.
The reference paper can be found here.
Time Series Forecasting of Amazon Stock Prices Using LSTM and GAN
This project uses Python to analyse Amazon Stock data. The feature extraction is done, and the ARIMA and Fourier series models are built in this regard. The long short term memory (LSTM) has been used with multiple features to predict stock prices. Along with that, generative adversarial networks (GAN) is used to predict prices on data fetched from an API as generator and CNN as a discriminator.
Find more information here.
ECG Anomaly Detection via Time Series Analysis
This project proposes a time series analysis dependent anomaly detection scheme. In this project, computers will be able to analyse real-time sensor data to identify any abnormal heartbeats. In case of an abnormality detected, the particular time series segment will be transmitted to the physician for taking appropriate action.
The reference study for this project can be found here.
Predicting Sports Popularity Using Time Series Analysis
This project is to measure the popularity of each league using the search data collected from Google Trends, which give real-time historical data on search words. With this project, it is also possible to compare and forecast how the sports league are trending with respect to each other using three models — trend plus seasonality regression, Holt-Winters Multiplicative (HWMM), and Seasonal Autoregressive Integrated Moving Average (SARIMA). Businesses interested in advertising or investing with either league may leverage these forecasts for deciding which sports league provides the greater or long-term value.
The reference study for this project can be found here.
Time Series Learning
This project aims at developing flexible methods that can fill in, backfill, and predict time-series using a large number of heterogeneous training datasets. In this case, the datasets we are considering consist of multivariate time series measurements. Each of these time-series has missing parts; the goal is to fill the missing input variables using deep neural networks or recurrent neural network solutions.
Visit here for more information.