Univariate Time Series Model for Ozone Concentration at Arosa Using TensorFlow

Template Credit: Adapted from a template made available by Dr. Jason Brownlee of Machine Learning Mastery.

SUMMARY: The project aims to construct a time series prediction model and document the end-to-end steps using a template. The Ozone Concentration at Arosa dataset is a univariate time series situation where we attempt to forecast future outcomes based on past data points.

INTRODUCTION: The problem is to forecast the monthly ozone concentration levels measured at Arosa, Switzerland. The dataset describes a time series of concentration levels (in Dobson unit or DU) over a 40 years period (1932-1971), and there are 480 observations. We used the first 80% of the observations for training and testing various models while holding back the remaining observations for validating the final model.

ANALYSIS: The baseline persistence model yielded an RMSE of 21.821. The CNN-LSTM model processed the same test data with an RMSE of 18.668, which was better than the baseline model as expected. In an earlier ARIMA modeling experiment, the best ARIMA model with non-seasonal order of (2, 0, 4) and seasonal order of (2, 0, 1, 12) processed the validation data with an RMSE of 16.6.

CONCLUSION: For this dataset, the TensorFlow CNN-LSTM model achieved an acceptable result, and we should consider using TensorFlow for further modeling.

Dataset Used: Monthly Ozone Concentration at Arosa January 1932 through December 1971.

Dataset ML Model: Time series forecast with numerical attribute.

Dataset Reference: Rob Hyndman and Yangzhuoran Yang (2018). tsdl: Time Series Data Library. v0.1.0. https://pkg.yangzhuoranyang./tsdl/.

The HTML formatted report can be found here on GitHub.