Time Series Forecasting - Which Algorithm can be applied

Question

Hi All,

I have a requirement for which I need to build a Time Series Forecasting. The objective of this requirement is to forecast the no of support tickets for the future.

I have a small amount of dataset with 3 months data where I have two columns namely date_time and the no of tickets created on each day. From this historical data we need to forecast what will be the forecasted projection for future tickets. Although the dataset is very small and out of this data we cant build any effecient model but I want to know the suitable algorithms and approaches for solving this dataset.

Once I loaded this dataset I visually plotted the data to see what are the time series components present in the dataset. But I cant able to identity any components such as Trend, Cyclicity,Seasonality so I need to know how can we proceed further.

I attach the image of the plot for your reference.  Any guidance and inputs will be of great help.

Thanks,

Karthik.

plot.PNG

DawnDuong · Answer

Hi @Karthik_7694

i don’t quite follow you. If there is no trend no pattern no logic there is no possibility to forecast — right?

Perhaps you should look into which factors drive the incident, form a hypothesis around the model first.

from the description alone, the number of errors are likely driven by factors such as number of users, introducing new software etc...

you first need ro understand what the input drivers are, then blend then with your “outcome” (ie the numbers of error reports) and then model out the relationship

Only then you can decide which method works.

it is first a statistical learning question and then it becomes an implementation question on Alteryx.

dawn

Karthik_7694 · Answer

Hi @DawnDuong

Thanks for your inputs I will have a look into it . As I mentioned I have a dataset which does not have any seasonality,cyclicity,or any trend in the dataset so can you tell me which algorithms can be suited for it because I went through the algorithms list for time series such as Moving Average, Exponential Smoothing, Arima,SARIMAX,Holts Winter model,LSTM..etc but these algorithms are best suited for a Univariate dataset which has clear indication of time series components but these algorithms are not best suited if the dataset has irregularity or random fluctuations..So it will be of great help if you can guide me on this use case.

I also tried multiple algorithms although it does not have any time series components and evaluvated its performance based on metrics such as MSE(Mean Square Error)

Thanks,

Karthik.

DawnDuong · Answer

Hi @Karthik_7694

Usually we do more than 1 forecasting methods then compare the scores.

There is a fairly robust tutorial/step-by-step guide on how to set up for different methods under the Learn section. some examples:

https://community.alteryx.com/t5/Interactive-Lessons/tkb-p/interactive-lessons/label-name/Time%20Series%20Forecasting

https://community.alteryx.com/t5/Interactive-Lessons/tkb-p/interactive-lessons/label-name/Predictive%20Modeling

https://community.alteryx.com/t5/Learning-Paths/Data-Science-Learning-Path/ta-p/504157