Hello Readers !

  Lots of job seekers want to become Data Scientist whereas it is difficult without right path. We start searching Google about Data Scientist then what we get result as finding baby in Earth and you do not know where that baby is. I mean to say that you get lots of link suggestion where few blogs will talk about technologies used by Data Scientist, few will start explaining statistics equations which scares lots of guys who do not know statistics or mathmatics and so on. You also find some degree in Data Science claiming tag industry expert training and charge huge fee as they say we have collaboration with so and so Indian or specially abroad university. Some people keep talking about Machine Learning as well. Why I am writing this because all these thing creates further confusion.

  Today we are going to discuss what is Time Series Forecasting? We will discuss everything related to Time Series forecasting and why it is required.

  Time Series forecasting is concept which helps you to forecast value for future based on previous trend of same type of value itself. Now you may think what does this same type of value itself means? Let's quickly try to understand this with one example of saving money. You are saving 1000 INR or USD or any other currency since last 5 years and you noticed that you have saved 60000 INR or USD so far. (Note: Your monthly savings of 1000 INR or USD will be called as Historical or previous data in Data Scientist's language. and Your data is at month level because you are saving money monthly). Now you want to check what you will save on monthly basis in next upcoming 12 months (This requirement will be called as forecasting monthly savings of next coming 12 months in Data Scientist's language). Forecasting savings for next 12 months will be done by Time Series Forecasting. and here we are going to forecast your savings only as we have your previous savings pattern. That is why I wrote earlier same type of value itself. We cannot predict savings of your brother or sister based on your savings pattern because we do not know what he/she is saving.

  Same way Data Scientists need to forecast various types of values using Time Series Forecasting like Sales/ Employees to be hired/ Employees will leave company/ Profit or more as per business requirement. If you have knowledge of Machine Learning then you may be thinking that Why we need to use Time Series Forecasting when we have Machine Learning?

  Let me first tell you Machine Learning and Time Series Forecasting both are just concepts or just a text name. Machine Learning or Time Series Forecast does nothing. We need to use the techniques available under the umbrella of Machine Learning and Time Series Forecasting. We have Regression analysis, Logistic Regression, SVM, Decision Tree, Random Forest and more in Machine Learning and these techniques solve regression problems and classification problems. Same way we have ARIMA, ARIMAX and other more types of techniques under the umbrella of Time Series Forecasting and Time Series Forecasting is used to forecast time based values and we have already discussed about savings example. You can go through our Machine Learning article, it will give knowledge about Machine Learning- Introduction to Machine Learning

You can also watch video of Machine Learning Introduction below.

  Let's try to understand Machine Learning and Time Series Forecasting based on example of our life. Suppose we have one concept of Food but food is nothing and then we have another sub concept of Eatable and Drinkable. Eatable cannot replace Drinkable things and Drinkable cannot replace Eatable things. Then further under the umbrella of Eatable we have Pizza, Burger, Vegitable, Cake, Sweets and more and Pizza cannot replace the taste or requirement of Cake. and Cake cannot replace Pizza. Same way we have Cold drink, Milk, Mango Shake, Banana Shake, Vodka, Rum under the umbrella of Drinkable. Each of these have their own significant. If person is looking for Rum and you give Milk then he will not be happy. Same way if somebody who does not drink Rum and you give Rum then he will not be happy.

  Same way we have Data Scientist (Prediction and Forecasting job role) and under the umbrella of Data Scientist we have further concepts of Machine Learning, Time Series Forecasting, NLP and more other. Then further under the umbrella of Machine Learning we have regression problem techniques like Regression anaysis and classification problem techniques such as Logistic Regressiona and Decision Tree. Same way under the umbrella of Time Series Forecasting, we have ARIMA, ARIMAX and more types of techniques to forecast values. You can consider Time Series Forecasting as drinkable and it is as import as drinkable is for you life. I have taken this example so that you understand the importance of Time Series Forecasting.

Type of Techniques in Time Series Forecasting?

  Now let's go back to example of savings. We were having savings details of 1000 INR or USD per month of last 5 years and we had to forecast savings for next 12 months. Here savings will be considered as variable and we want to predict savings only so it will be called as dependent or output variable as well. As we have only savings, it is single variable and we need to use ARIMA to forecast single variable future value. ARIMA stands for Auto Regressive Integrated Moving Average. 

  Now let's discuss another example. Your savings depends on your salary, your expenditure and your other income sources as well and when you started saving your money, you will be having some salary, some expenditure and other income sources if any. Now we will be having 4 types of data Savings, Salary, Expenditure and Income from other sources. Here your savings depends on all other 3 data - Salary, Expenditure and Income from other sources. Therefore your savings will be called as Dependent Variable in the language of Data Scientist and your Salary, Expenditure and Income from Other sources will be called as Indepent variables because your savings does not impact Salary, Expenditure and Income from other sources. As per requirement, you want to forecast next 12 months savings considering you Salary, Expenditure and Income from other sources. Here we will have to build ARIMAX model or we can say we will have to build multi-variate Time Series Forecasting model. 

  Same way suppose you get bonus or variable payout from your company half yearly like Jan and Jul of each year. If we have this data as well then this scenario will be called as variable payout and if we include this data as well to forecast 12 months savings then it will be called as half yearly seasonal model because when ever you will get bonus or variable payout you will be able to save more compared to normal months. This kind of problems are solved by Seasonal ARIMA or Seasonal ARIMAX models.

What technology required to build Time Series Model?

  Time Series Forecasting models can be build in SAS, R or Python. We have more other software solutions as well but you need to learn that software which is asked by companies because you are looking for job and you need to learn whatever organizations are looking for. In future, we may have requirement of another technologies like Julia is evolving as one of data science language. Once you learn Time Series Forecasting then concepts will be same but you have have to keep updating yourself for technology as per company requirement. You can have quick look on SAS, R and Python as well. 

Introduction SAS, R and Python

Which is best for Data Science - SAS, R or Python?

Where will you get job after learning Time Series Forecasting?

  Time Series Forecasting is used by Data Scientist job role guys. All those companies require Data Scientists which are looking for to run their business effectively and wants to make profit. Why I said this because you will be able to understand importance of Data Scientists that all companies need Data Scientist.

You can have quick look on job roles provided by organizations. Data Analyst vs Data Scientist vs Data Engineer vs Data Visualization

Can non statistics background guy learn Time Series Forecasting?

  Yes you can learn Time Series Forecasting whether you have statistics knowledge or not. It depends on your instructor. You will not have to build model by your hand. Time Series models will be build by SAS, Python or R. You need to learn how to operate SAS, R and Python to build Time Series Forecasting model and once model results are out from SAS, R and Python then you must be able to understand those results and implement to business.

You can also have quick look on video tutorial of Introduction to Time Series Forecasting.

Let me know in case clarification required. You will have to register on site to add comment in case you want to ask question.