I have a dataset that contains (date of request , date of flight, from, to and price), now I want to get (date , from, to) and then predict the number of flights with these features by machine learning and data mining techniques.
What's your idea for predictions?
You can use your data at hand as labeled data then train a neural network. This will give you an idea how to start. If the ML cam be simple, never get too complicated just for fancy.
Data mining is not magic, it only can calculate faster and only according to the algorithm given. So you will have to help by preparing your data with some aggregations and provide training data usually about 30%(for instance, if you want to know the prediction of the next 2 years, you must have this year real data; to train the prediction model). You will also have to know the algorithms that can solve your problem, because some only work in specific situations or goals. In your case, why not have week days instead of date of flight, or number of days from request date to flight date, or months if you have data of one year at least and so on. Like this it will produce better information, for example weekends usually results in higher number of flights than week days, or certain months of the year have higher number of flights and so on. Depending on your data ranges and purpose of the study you can prepare your data. Sometimes these modifications in variables can only be determined by several trials until you get higher prediction rates.
Hi,
Your data have to contain the "number of flights" at the very early stage. Once you have it, you can forecast it using algorithms like Linear Regression (with the use of proper Lag variables).
HTH.
Dr. Samer Sarsam
Dear Ruhallah Ahmadian ,
Predicting flight delays using ML is a very hot topic these days. Your problem, predicting number of flights, is one of its simpler forms. I agree with Abdallah Salim Al Mazidi . The data set you use needs some modification such that one variable (or feature) can show the number of flights during a fixed time like one week or one month. By training your classifier using a portion of such data, you can predict the number of flights for unseen weeks (or month,...) as well.
Hope it would help!
Your question: How can I predict that how many flights will be there ?
My humble opinion:
1. The most simple way is to retrieve databases from most airlines available at that city airport.
2. Not all airlines are available on a partcular date (weekdays), you may adopt data mining to find our a pattern of available flights on that specific weekday during a
past period (e.g., last 3 years).
3. The you may apply ML techique(s) to learn and predict how many flights will be available on your chosen date.
Thnak.
Dear Mahdieh Zabihimayvan and Abdallah Salim Al Mazidi
I aggregate my data on (from_id, to_id, data) ، now I know the number of flights for each (from_id , to_id, date), and then I split the data to train & test and then forecast the test by Prophet (it's time series forecasting https://github.com/facebook/prophet) but my score is negative. you can see my result below , do you think is it a good prediction or I should work more on data?
In my experience you should try with Poisson or Quasipoisson mode, taking into account that the output variable is discrete. Also, a Random Forest may be a good option, applying a cross-validation approach to fit the parameters of the model.
Hello,
I completely agree with Dr. Samer Sarsam . You need to look at an evaluation metric to evaluate the performance accurately.
Best,
Dr. Fernando
- Badges
- Science topic
- Similar topics
- Computer Science
- Data Mining