I will help create forecasts based on time series data
Data Scientist, Consultant
About this Gig
Welcome to my gig for time series analysis and forecasting!
Time series is a set of data points collected for the same entities over an extended period of time. Eg. Daily temperature recordings for a city for a 5 year period.
Time series data tends to have certain characteristics like trend, seasonality etc. Understanding the behavior of the data in terms of these characteristics is imperative for time series analysis.
Forecasts prepared using time series methods can be accurate and can factor in a variety of real-world phenomena & their impact on the way the series varies over time.
I follow a structured process to develop forecasts based on time series data:
- Data preparation: Organize and clean historic data to bring it to a suitable format for analysis
- Exploratory data analysis: Comprehensive analysis of the dataset to identify underlying trends, seasonality and other latent patterns
- Model selection: (ARIMA, ETS, LSTM, HW)
- Model development: Develop a robust model to capture complexities in the data
- Forecast generation: Create accurate and reliable forecasts to support decision making
- Model evaluation: Evaluate model performance through accuracy metrics
- Visualization
Programming language:
Python
•
R
•
SQL
Frameworks:
Scikit-learn
•
Keras
•
Panda
Tools:
Jupyter Notebook
•
TensorFlow
•
Excel
•
Colab
•
Other
Other Data Science & ML Services I Offer
FAQ
How much historic data is required for time series forecasting?
A minimum of 24 months of data provides a good view into the seasonality and trends. Taking into account at least 2 seasons, out-of-time validation and meaningful data, it is advisable to have at least 30 data points in the series that cover at least 2 seasonal cycles.
How accurate are the forecasts?
Time series forecasting methods can provide extremely accurate forecasts (< 5% error) in cases when the data is exceptionally strong. However, depending on the case, models with error rates below 15% are still considered to be strong forecasting models.
