ARIMA Model In Python| Time Series Forecasting #6|

[@NachiketaHebbar]

There is a small correction in the plot. By mistake I had trained the model on the entire data set, instead of just the training set. While shooting the video, I noticed this mistake and made the correction, however I forgot to rerun the code. On fitting the model on training set, the plot that you would probably get is a somewhat constant plot that ranges between the values of 44 to 46. That is fine, it just means that the model would have got a lower error in forecasting around the mean value, instead of fitting to the irregular variations. You can also try with a bigger data set, or other models like random forest or even RNN's.

[@amitajoshi916]

Are you referring to the following? start=len(train) end=len(train)+len(test)-1

[@anuragsen8421]

@@amitajoshi916 No he referred to the part where he trained the model. he trained it using the whole data

[@nilsloeken]

So could you give me the correct code in this case?

[@ramirotapia2217]

and the auto_arima as well

[@nextReanimation]

Hi. I tried it with my own data but my prediction curve is very static. Any idea why this could be like that?

[@divyanshuchaudhari5416]

Best Video on ARIMA on youtube handsdown.

[@renatorodrigues2686]

Great stuff, one of the simplest arima tutorials out there. Great for beginners!!!! Keep up the good work!

[@MANVITHAH-yi1zh]

Very Neat Explaination with Great effort... Thank You in the same way

[@kalyanadepu5666]

Theoretical concepts of Acf and pacf matched with practical Acf and Pacf . Thank you 🎉

[@spaghettiplants]

Thank you for this! It's a great help and it helped me understand how to implement an ARIMA model, specifically in deciding the order of the AR and MA components.

[@raghavverma120]

You can do this manually too by looking at acf and pacf of resduals but then autoarima is always handy..

[@Shruthioffice]

It is explained clearly and well articulated with short time.

[@sriadityab4794]

I also follow many other ML channels but yours is the best one. Keep rocking bro 🤗

[@NachiketaHebbar]

Means a lot, thanks!

[@newmanokereafor2368]

Just like I wanted to have it explained. This is so good. Thank you @Nachiketa Hebbar

[@akankshakoshti9168]

Your videos on time series are cool. Easy to understand and on point. It would be great if you post a video on the rolling forecast.

[@pawanpatil5285]

bro, temp dataset is seasonal dataset.. u can see it in ploting as well.. u have to use sarmia for that.

[@prosimulate]

You have a new subscriber😊 Rather slick Nachiketa, well done 👏

[@fredbennett5285]

Thank you for the clear explanation

[@rahulghosh4289]

In a nutshell you have to watch all older videos before going through this video.

[@sangeethasaga]

First time am understanding a time series video in the first view

[@kilaa0007]

thank you, you saved my final year project

[@mishradwait]

Can you please share your project details? Please

[@Matthew8473]

This is top-tier. I read a book with similar content, and it was top-tier. "The Art of Meaningful Relationships in the 21st Century" by Leo Flint

[@zaheerbeg4810]

Nachiketa, You are a nice teacher man, keep posting please

[@alexlefavi6943]

Hi Nachi, having a lot of fun with this video and data. I am working with the same df and code, simply trying to replicate your results. I get stuck when I print out predictions (7:00 in video). I have much less variation in my predictions. numbers start at 44 and eventually stay on 46, so my plot is a straight line. Everything before this has matched: p value, order, aic Any idea what I am doing wrong here? Thanks!

[@piratetechie2411]

Same thing happened with me as well.

[@alexlefavi6943]

@@piratetechie2411 yea i eventually figured it out. Can't exactly remember, but I think he had been running his arima model on the entire datafile, not the train, which is what I had the impression of. Mess around with step 5:55. If you use the train you'll get the same stats he shows. Then try that step with entire df instead of train. You'll get slightly different stats but i think the charts toward the bottom will look better. Let me know if you get it. This gave me a lot of trouble

[@SaikatGhosh752]

@@alexlefavi6943 can you help me solve the problem . I am too facing the same problem and cant find any solution to it

[@tengkuaimi6354]

@@alexlefavi6943 can you help me, how to make a coding to change from entire datafile to train datafile only?

[@francoc7698]

@@alexlefavi6943 hey sorry I know this has been a long time but I encountered the same issue... mine predicted table shows predicted_mean... and it is a constant line.. do you know how I can fix it? I tried to use "model=ARIMA(train['AvgTemp'], order=(1,0,5)) but it still shows a constant line...

[@jos6708]

getting error like this how to solve this... NotImplementedError: statsmodels.tsa.arima_model.ARMA and statsmodels.tsa.arima_model.ARIMA have been removed in favor of statsmodels.tsa.arima.model.ARIMA (note the . between arima and model) and statsmodels.tsa.SARIMAX. statsmodels.tsa.arima.model.ARIMA makes use of the statespace framework and is both well tested and maintained. It also offers alternative specialized parameter estimators.

[@hrithikhm808]

I'm getting the same error 👆

[@asimhassan28]

me to

[@nasosskaltsas9890]

write: sm.tsa.ARIMA(...)

[@shadyizloo]

i am not getting it to do forecasting for future values. can you assist on what needs to change for it do forecasting for future dates? what needs to be changed?

[@SanghamitraOjha-g7j]

Hello sir , I couldn't able to downloads the dataset , can u pls kindly give me the link of the data set

[@KKDEV-cc4xd]

This is Great!, Thanks

[@doncharles3355]

I watched this video. This was very good and well explained. I am familiar with Matlab, but new to Python. Nevertheless, I was able to follow this. My data set was different, but in the end the code worked. Keep up the good work!

[@vishalgarg7207]

In your case the data was stationery but could you please recommend what are the best approached to stationerized the data ? If in DF test the 1%,5%,10% is greater then ADF?

[@apoorvabharti9907]

Hi I am working on a project where i have to predict the registration percentage drop. I am retrieving the data from an API. But the accuracy is too low, the predicted mean graph is extremely inaccurate in comparison to the actual graph, could you maybe check and help me with the code?

[@hasantahir3926]

@nachiketa question - Is it possible to take into account multiple inputs? How? Also, if you have seasonality? How do you use SARIMAX?

[@eneseren8332]

You are awesome. Thanks for the tutorial.

[@rushabhkankariya4769]

I am facing error at model "prediction must have end after start" how should I fix it?

[@ruchiradhongde5614]

Hi,Can we fit ARIMA model on multivariate data?(2-3 independent variables)?

[@iamadam9743]

I had same doubt

[@chaddruck8694]

if I made my data stationary, how can I get my predictions to reflect actual values instead of the decomposed data?

[@tanmoyde2268]

Hi, i ran the ARIMA model . everything went fine, except the graph visualization was erratic. please comment . it will be great help

[@harshitgupta5053]

instead of date if we have time in seconds what to do?

[@mingkwan5280]

do we just put all this code in python ide and run? I haven't learn programming but I need to use it to forecast now..

[@peacemaker_69771]

Hi Nachiketa, In 5.48 of the video the order you are mentioned doesn't work for me. I used the same ipynb file you have used. What should I do ??

[@hanS-ti1jg]

Help, statsmodels.tsa.arima_model has been remove, what to do?

[@harsharajendran41088]

If the values are in datetime format how do you write the index_col and parse for it? with the values increasing every hour.

[@keshavdutta2503]

hey i am stuck at make prediction on training test as i am not able to run that because predict() type is not found is occuring . so would you suggest how to solve this. It would be very helpful

[@kevinalejandro3121]

If I only want to consider a certain lag in my ARIMA model?? for example only consider the lag 3 but i don't want the lag 1 and 2 in my model, How can i do that ??

[@NikolayRein]

unfortunately I can not find the link to dataset

[@TusharBansal-w2n]

Hi Nachiketa, I have a question. In ARIMA model the integrated part allows us to difference the time series to get a constant mean. This will remove stationarity only in cases where the series violates only the non-constant mean property. But if there is a series which has volatility and seasonality then what can be done in such a scenario?

[@hasankaracayir9015]

Hi Nachiketa! I've a question. i want to predict x with values of x, y and z. With ARIMA, i can predict x with historical values of x, but can i include the historical values of y and z aswell to predict x? Is there a parameter?

[@simrangoswami06]

Hi, In fitting a SARIMA model, I got the RMSE = 0.4. Could you please guide me how can I comment on the percent accuracy of the model based on this?

[@harsharajendran41088]

I tried the forecasting for a dataset comprising of 25000 values. Initially I had kept the training set up to the last -200 values. The fit was good. But later I tried it for half the data set. I took the training as 10000 values. The fitting didn't go well. Later I tried fitting again with up to -200 values and it still wouldn't fit. WHat should I do now?

[@radhika204]

very good video.Nicely explained

[@MrPranavsai]

how to handle seasonality? what if the number of data points is less than 50?

[@nina42703]

thanks for great video. I have a question. I applied this code to my data auto arima. The only difference is that I have seasonality=12 months. So how should it be the code for manual arima?

[@chillwithme798]

i can not predict, that error is Cannot cast ufunc 'subtract' output from dtype('float64') to dtype('int64') with casting rule 'same_kind', pls

[@kevinalejandro3121]

¿Which is better an ARIMA model or a Neural Network? ¿or sometimes one is better than the other?

[@NachiketaHebbar]

It depends on the use case. But a neural network is capable of capturing way more complex relationships present in time series. However it would be a waste of resources when say, maybe a simple Arima model of order (1,1,1) could do the job.

[@chetanmundhe7899]

very nice explaination...keep posting

[@NachiketaHebbar]

Thanks!

[@billzalokostas5811]

Does your ARIMA model overfit? I am asking because I observe that the predictions is just like the actual values shifted by 1. Why does this happen? Thank you in advance

[@shahrizalmuhammadabdillah3127]

my model always stop or error in model = ARIMA(train blabla bla i dont know how to fix it

[@Rg20107]

in task if i have two csv file in TSF then how could i find best model from that two csv file ?

[@pointblank722]

Can you build an algorithm which detects anomalities in time series data by predicting future values and comparing predicted future values and comparing them with the real values??? This came to my mind and maybe someone thought this before..If yes, I would love to see the code...

[@josephsevere238]

You can probably do this using p-values. P- values basically an indicator that detects the likelihood of an event. In this case, anything lower than a p-value of 0.05 would be susceptible.

[@22sohail22]

how to make prediction for non-stationary data? OR do we have to convert non-stationary data into stationary first?

[@cyrilmouttet1856]

Hi Nachiketa. Thank you for the video. I'm quite confused since at 7:23 (kzbin.info/www/bejne/bnemdaN8nclksM0) you are stating that the model is performing pretty good. However there seems to be one lag between the actual and predicted values, which in my opinion tends to a pretty bad model. I would expect the predicted data being exactly over the actual values for a good model. Or did I miss someting there?

[@harsharajendran41088]

Do a video for support vector machine model as well. Especially where the F-Score is calculated between two datasets having the same column names but different values due to the various conditions or parameters they are subjected to.

[@saketsrivastava84]

Very good video 👍🏻

[@christiankyleroldan558]

The prediction values were 30 months only. Is there any ways I can predict for at least 36 months ore more?

[@francoc7698]

hey Nachiketa, I am using your method to build an ARIMA model to predict a product balance. But when I plot the pred=model.predict(start=start,end=end, typ=levels) line, mine is showing a predicted mean value... meaning all the predicted values are the same and at the bottom of the predicted value, it indicates "Name: predicted_mean" dtype:float64"... Do you know why this is the okay? How can I avoid its predicting the means instead of actual values? Thanks!

[@adedolapoogungbire7088]

Experienced the same. Kindly let me know how you were able to resolve this.

[@Lejik007]

hi, how do you predict next day, or next few days(steps)?

[@shadyizloo]

have you got an answer to this i am also looking to do this but there is no response

[@Lejik007]

@@shadyizloo i did something like this with NN

[@shadyizloo]

@@Lejik007 something like what

[@beyondai268]

Hey , My RMSE value is negative , what is meaning of that ?

[@abhigyan2619]

how bro, can you show the whole process manually? I mean mathematically without code?

[@bumpagab]

Thank you! Very helpful.

[@Dhiya_In_Korea]

Hello brother. I used my lab's greenhouse environmental data. my main goal is to find dewpoint. it shows an error while training the model part. can you please help me to resolve the error? my error is "NotImplementedError: statsmodels.tsa.arima_model.ARMA and statsmodels.tsa.arima_model.ARIMA have been removed in favor of statsmodels.tsa.arima.model.ARIMA (note the . between arima and model) and statsmodels.tsa.SARIMAX. statsmodels.tsa.arima.model.ARIMA makes use of the statespace framework and is both well tested and maintained. It also offers alternative specialized parameter estimators.

[@hamsavardhinim1278]

Great video.

[@parthbisht5597]

Bro after applying the adfuller test to the predictive variable it shows the error if tolerance is not none please help

[@manglem10]

What if the index is only year instead of day month year? My dataset is like Mean Temperature data Year. Jan Feb. March ....dec 1969. 6. 8. 10. 2 1970. 7. 7.5. 8.5. 3.5 1971. 3. 6.5. 6.5. 4.5 ... 2000 In such case how do i prepare the dataset for arima analysis to forecast monthly mean temperature values ? Will be grateful if you answer Thank you

[@ashishtinker8119]

Namaste, me Ashish Tinker, Jaipur se. Mujhe ek help mil skati he kya please. Me jab apka arima model use kar rha tha to shape of data nahi ata or excel file jise read kar rahe hein usme bhi error ata he , mujhe kya karna chahiye

[@bryancapulong147]

auto_arima gave me arima(0,1,1) and the prediction came out as a slightly diagonal line... how bad is it?

[@bryancapulong147]

The prediction was following the trend of the dataset in the beginning until in the end the plot just became the negative version of the actuals

[@rabbitazteca23]

Hi! If my dataset has seasonality in it, can I use seasonal=False as a parameter for the auto_ARIMA? if my model has seasonal component, do I have to set this parameter to false?

[@raghavverma120]

Remove the seasonality of the model.. by adding ‘d’ component..

[@raghavverma120]

You need to make the distribution non stationary only then you can fit the model.. best way to do it is by taking difference ..

[@dhavalchauhan]

Hello....I have collected Data of 26 respondents for for 42 days. So I have 42 variables for each respondents. Which means 42 values of 1 respondent for 42 days .... 46×26 rows and 42 columns... So how can I fit ARIMA MODEL in my data ? Is it valid for arima to fit on average value day (42 days) wise of all 26 response

[@mustafamehdi594]

Hey. Can someone please guide me why it's giving the error "year 0 is out of range" while iam printing my predicted values?

[@marciokoko1]

What to do when the suggested order is 0,0,0 and the predicted values are identical?

[@anweshgandham6776]

Thanks for simple explanation very useful . For my data using ARIMA ...mean and rmse are in similar range .what to do in such cases .

[@KRISHNAPRASAD-xy7tr]

Thanks for clearing my doubts on this alg, I am currently working on utility usage forecasting, my uniqueness can be on date & service point id, so shall i create index on both columns?

[@MrAkshayNR]

what if we have non stationary and what to do next

[@ankitbagga5713]

don't we need to make it stationary before pmdarima?

[@dicloniusN35]

how to transform data to stationary and fit it into arima ? than get predictions for this data withot transformations

[@harryfeng4199]

Can I use it if the data is not stationary??

[@hanS-ti1jg]

What do you do if your data is not stationary?

[@driani65]

very useful tvm!!!

[@christenthomas434]

Hi Nachiketa thanks for the video, I am looking on a time-series data to predict the infection rate for covid. But as you said arima model can be used on stationary data, any suggestions on how I should approach this ?

[@siddhijain3802]

if we have NA values in the dataset..but we dont want to drop them..what is the best method to fill them? like should we take average or median values to fill just like in normal regression models..or are there any better way? PLEASE help!

[@NachiketaHebbar]

There is no hard and fast rule to determine which method is gonna work best for you. Mean, median and mode are the most common methods used normally to fill in missing data. I suggest you practically try all three to see which method gives the most accurate results for you. There is also another method called tsfill which i suggest you google, it fills in missing time series data using interpolation.

[@swarupkumarmondal3323]

Hey, can we train arima by taking two columns and comparing between them? What are parameters to do this if possible?

[@dp_9290]

arima is univariate model, so 2 col is not practical

[@himeshkoli8607]

Dude ARIMA can handle non-stationarity right?? So after differencing if my data is still non stationary so should I fed that 1st order degree difference to ARIMA or should I directly fed the original data (without differencing) to ARIMA??

[@bhaskarbirla6928]

Hey! Thanks for the video. Its really helpful. Want to confirm if in Augumented Dickey Fuller test, the null hypothesis is Data is not stationary. If p-value0.5. Please correct me if I'm wrong. Thanks again :)

[@azri8799]

no, we reject h0 if p value less than 0.05

[@rishikeshks539]

How to access and store the coefficients of an ARIMA MODEL into a numpy array

[@oliverxu7027]

The model looks not great right? It’s always lagging one behind and not able to keep up. You should look at the rsquared

[@amitajoshi916]

Great video - thank you! Are you able to paste the code for the correction you made? Also maybe share a copy of the entire code with data. Thanks

[@ZibaFact]

Plz make on deployment also

[@abhishekkusre4202]

nicely explained bro!

[@SalehaBegum-t8h]

you should provide the link of the dataset that you are using i

[@khanrubayet4092]

It is possible to predict 30-50 years temperature prediction by using ARIMA model

[@brianlivian3149]

Great vid

[@highbrassrule]

Thank you friend, this saved me

[@NachiketaHebbar]

Glad :)

[@vandhanasankar8389]

Could you please share the GitHub link or jupyter notebook link? Thanks!!

[@saranbodduluri2571]

Sir, can time series forecasting be applied to the percentage change of the closing prices of stocks?

[@saranbodduluri2571]

like can it be used to predict future retruns based on past returns?

[@yousif_alyousifi]

How to used ARIMA if we have 5 variables? For example, Y= sales X1=TV, X2=Radio, X3= newspaper, X4=FB, and X5=youtube

[@niharikasharma8486]

Hello Nachiketa. When I am running that future date section code , that ( index_future_dates..) line , this section is giving me following error . . value error : Length mismatch : Expected axis has 31 elements, new values have 325 elements. I hope you will help me in solving this error ..

[@ngbx8477]

Change your start and end to match 31 days. I did that and it worked.

[@manojrangera1580]

i think retraining the model is not good for our model. because we train our model on training data and and see how good our model is performing on test data then after we need to use this model to make future prediction. if we retrain our model of full dataset then it is not good, because how will we find out that our new model is performing good ya not so, it is not good to retrain the model.