Prediction of full cycle using partial cycle data

The cycle starts with 100% ends at 0 and is dependent on time and some other parameters. I have initial cycle data from 100% to 80% and by using this data I want to predict remaining data. Which algorithm should I use to predict the remaining data.

[TED MOSBY]

What exactly do you mean by cycle? Please share specific details about your model and what is does and what are you trying to achieve along with the dataset you are using to be able to help you further.

Answer 1

[Ayush Anand]

Hi,

Predicting the remaining cycle data when you already have data for the initial part of the cycle can be approached through various time series forecasting and regression techniques. Here are some you could try out:

1. Linear or Polynomial Regression:If the relationship between the cycle percentage and time (or other parameters) can be modeled linearly or as a polynomial, linear polynomial regression can be a good choice.
2. Time Series Forecasting Models: If your data is time-dependent, considering time series forecasting models could be beneficial:

• ARIMA (AutoRegressive Integrated Moving Average): ARIMA combines autoregression(effect of past observations), integration(to capture the general trend), and a moving average(to penalize errors) and is an effective algorithm that can be used for prediction. (Read more about ARIMA here: https://www.mathworks.com/help/econ/arima.html )
• Exponential Smoothing (Holt-Winters): This applies smoothing to capture trends in the data .( Read more about Holt-Winters here: https://www.mathworks.com/matlabcentral/fileexchange/53699-holtwinters-y-l-m )

3. Other Machine Learning Models:

• Gradient Boosting Machines (e.g., XGBoost, LightGBM): These build models in a stage-wise fashion and are good at handling various types of data, including nonlinear relationships.
• Neural Networks: If your dataset is large and complex, deep learning models like LSTM (Long Short-Term Memory) networks or GRU (Gated Recurrent Units) can capture patterns effectively, especially in sequential data. These are particularly useful if the cycle's progression depends on long-term dependencies.

Hope this helps!