Please find the sample dataset attached.
Here, as we can observe there is a forecast and actual data. However, we have only for about 8 years (2000- 2008) of data.
Trying: XGBoost, ARIMA
Challenges facing:
XGBoost (**Can this be used for time series analysis**? because it considers all parameters as it is not even time)
ARIMA (Not sure how to choose p,q,d for this particular dataset)
Flexible with both R/Python
Could you please suggest us which algorithm would forecast the next 8 months with considerable accuracy ?
Am not sure if XGBoost can be even applied for time series, please share link if this is practical
Also, since we have about 8 years of data on monthly level, Neural network such as RNN/LSTM might even work ?
Hello Ian,
Talking about software: ARIMA will work on both R and Python. If you want to use machine learning, you will have to use Python with keras/ tensorflow.
About your time series: 1) It seems really weird, I am doubtful if any of the techniques will be useful for you. 2) If the attached dataset if the actual time series that you want to predict, you must first process your data to have the entry for missing values '?. 3) What are the red values?
You need 8 months ahead forecast with 8 years of data. So, your problem is suitable for machine learning algorithms. RNN is outdated. You should try LSTM, if you want to try neural networks.
Given the weird pattern of your data, I would recommend that you first try out ARIMA and then LSTM if needed. Compare forecasts against a persistence model to check performance of ARIMA and LSTM.
It appears that you are not very familiar with either ARIMA or LSTM. So, you should go through some tutorials first, to clarify your concepts. Following resources will answer your question regarding ARIMA:
1) https://www.youtube.com/watch?v=bqvZL8Ww3aA (Python)
2) https://www.youtube.com/watch?v=OvkOG8TNxFE (R)
3) https://stats.stackexchange.com/questions/44992/what-are-the-values-p-d-q-in-arima
Following tutorial will help you for LSTM:
https://machinelearningmastery.com/how-to-develop-lstm-models-for-multi-step-time-series-forecasting-of-household-power-consumption/
Good Luck!
If you have large no. of dataset then use Deep Learning otherwise use simple ML techniques. For forecasting ML algorithms are different from classification. Use ML algorithms like: X-Boost Regression, Gradient Boosting Regression, Decision Tree Regression, SVR etc.
If your data looks exactly like this sample, there is no way of using any algorithm. The reason is obvious, there is no proper problem statement. What kind of problem you wanted to solve using this data?
As I can understand, the 'Forecast' column is the output of a model that predicts some value, and the 'Actual' column is the reference values to validate the mentioned model performance. So, using this data you can quantify the quality of the model (which is already available) by RMSE or other parameter. I do not see any other purpose of this dataset.
- Badges
- Science topic