Dear Experts,
I want to make a predictive model with the problem classifications (two classes: 0 and 1). Therefore, it would be mind to receive a suggestion from you about any method in Artificial Intelligences (AIs) and others. Note that, for example, my target problems vary regarding the time variations and etc. I hope that you all will be kind in my attempt.
Thank you in advance for your help.
Gratefully,
Phimmasone
Dear Phimmasone Phim Thammavongsy ,
I would say that the answer depends on your input data type and size.
If you are dealing with time series and you have enough training data, I would suggest using deep learning models such as recurrent neural networks (RNN), long short-term memory (LSTM, which is an improved version of RNN), or convolutional neural networks (CNN).
If you are dealing with features already extracted, and only some of them are related to changes in time, you may try to use classical machine learning approaches, such as k-nearest neighbors (kNN), support vector machine (SVM), random forest (RF) or shallow artificial neural networks (ANN).
In this case, one model will perform better than the others based on the distribution of your training data (i.e. if class are linearly separable, then SVM will perform great; if the decision boundary is more complex than linear, ANN will outperform the others; if the decision boundary is not a single one, but many of them, kNN would provide the best results).
Dear Phimmasone Phim Thammavongsy ,
You can Use Recurrent neural network for your model, and use LSTM which is one of the best in this area.
Dear Luigi Borzí, first of all, I would like to say thanks for your useful answers. I am playing with the 90,000 samples of 6 features.
Dear Amin Honarmandi Shandiz , it is a great suggestion for me. I am now considering the LSTM. Thanks a lot for your answer.
Dear Kenan Morani
, your answer made me more clear and confident. I am now considering deep learning as the LSTM which agrees with Amin Honarmandi Shandiz and Luigi Borzí . Thank you all so much for your answers and appreciate your contributions.
Phimmasone Phim Thammavongsy i hope it will be useful for you https://www.tensorflow.org/tutorials/structured_data/time_series
you can use any classification ml-model but you need a lot of data for get high accuracy, and in time series forecasting the windowed dataset creating is very important
Leonid Kupershtein Thank you for your recommendation. I am interested in windowed dataset creation. I have gotten a confusion about it. Normally, I have a 2-dimensional dataset as time changes (seasonal and diurnal variations) and data attributes (its values at each time). Thus, what you meant by windowed dataset creating refers to the third dimension which splits the entire time series into subwindows based on the size of the given window. Is it right? If no, it would be great for seeing additional help from you.
Actually, there is no specific model known for always performing well . You might make your hands durty trying various methods. If you have just few samples, I suggest you to try classical machine learning algorithms (https://scikit-learn.org/stable/supervised_learning.html). If you have enough data, I guess Bi-LSTM might perform perform well for your time serie classification task. You can check this https://medium.com/@raghavaggarwal0089/bi-lstm-bc3d68da8bd0 to know more about Bi-LSTM. You can also use NAS (Neural Architecture Search ) method which aims to automatically find a best model architecture for a targeted task. seeConference Paper Neural Architecture Search for Time Series Classification
for more details.
You may take hourly based pollution control dataset and apply any one of the Deep Neural Network algorithm to develop a model for predicting any one/more pollutant(s). But to apply DNN, you require huge dataset.
A good illustration using R code is available om handling time series based classification modelling. 1) https://www.kdnuggets.com/2020/03/time-series-classification-synthetic-real-financial-time-series.html 2) https://machinelearningmastery.com/random-forest-for-time-series-forecasting/ 3) https://www.analyticsvidhya.com/blog/2019/01/introduction-time-series-classification/
Thank you very much Manojit Chattopadhyay for your answers.
- Badges
- Science topic
- Similar topics
- Computer Science and Engineering
- Artificial Intelligence