I want to built a binary classifier machine learning model. I want to use the model’s previous predictions as features for the future predictions, to take into account that my training samples are not independent.
Is there a framework to achieve this with scikit-learn, or any other python ML library?
I know this problem could be solved with memory-based Neural Networks architectures, like RNNs, LSTMs, …, but I would want to consider non deep learning approaches first (typically with gradient boosting models, tree based models, …)
Advertisement
Answer
There are several statistical models which you can use for predicting the future based on the past without using deep learning; in other words, for time series forecasting.
For example, you could use ARIMA/SARIMA/SARIMAX or VAR, which are statistical models which can be used for prediction.
You can consult the following links as starting points:
- https://machinelearningmastery.com/arima-for-time-series-forecasting-with-python/
- https://www.machinelearningplus.com/time-series/vector-autoregression-examples-python/
Also, do not forget to delve deep into the statsmodel python library.