Understanding the basics of Time-Series Data
Dataset used: AIRLINES PASSENGERS DATASET
We have a Monthly count of the number of passengers of an airline. Data is collected every month from 1949 to 1960. The time series dataset is assumed as a combination of trend, seasonality, and noise.
Data = Xtrend + Xseasonality + Xresidue
Trend – On decomposition, We can see the trend going upward in the long term with a slight flattening of the curve around 1954 and 1958.
Seasonality – If the data resembles some pattern over time, it is studied under seasonality. It occurs due to the influence of weather, vacation, holidays, and domain-specific facts. For example, in the Airlines dataset, The number of passengers was observed more around May every year and less in months around October. This can help to regulate the number of booking servers according to month. Bandwidth Optimization helps in the budget.
Noise – Shift of data from the sum of Trend and Seasonality is residue/noise. Noise helps the user to track the growth as more than expected or less than expected. We can see below that for 1954-1956 the number of passengers was almost similar to what was expected.
Techniques used in Time-Series forecasting:
- Time Series forecasting methods include Autoregressive Integrated Moving Average(ARIMA), Holt-Winter’s Exponential Smoothing, Facebook Prophet, etc.
- Time Series classification can be done using Recurrent Neural Networks(RNN), Support Vector Machine(SVM), Supervised Probabilistic Models like Logistic Regression, etc.
Applications of Time-Series in different Businesses:
- Forecasting Traffic :
A sensor at a traffic signal post counts the number of vehicles crossing the post every 20 minutes. With the data being collected, it forecasts the number of vehicle counts that will be crossing the post and thus predicts the traffic at the post. This type of forecasting can be highly used to avoid traffic congestion, de – routing people, traffic distribution, and efficient management of traffic poll booths.
- Forecasting User Spending Habits :
By studying the past data of user spending habits of an e-commerce website. The website will be pre-ready with product stocks, server availability for the period in the year when we see a spike in user spending habits. This kind of forecasting is highly useful in periods of festivals, sales, etc.
- Forecasting Customer Satisfaction :
Customer satisfaction is generally measured by the customer sentiments calculated from the reviews, feedback, etc. A company is all about its customers.
If we are measuring customer satisfaction in the range of -1 to 1, we can keep a check on the trend of this value. Suppose there is a downward trend, our model can predict by what time the customer satisfaction can go below a threshold value and accordingly, the company can make changes in strategic planning and decision – making to retain the customers.
- Forecasting Employee Churning :
Attrition and new employee hiring is a very important decision for many companies. All the companies want to have less attrition and hence less cost and efforts in new employee hiring. Forecasting helps in checking the attrition rate and accordingly manage the new employee hiring.
- Stocks Prediction :
Stock price values can be treated as a discrete-time series instead of randomly generated values. It can be seen as a set of well-defined numerical data items collected at successive points at regular intervals of time. Using models like ARIMA, SARIMA, SARIMAX, FBPROPHET, LSTMs give more authentic and reliable results then direct forecasting. These models study the past data as input and forecast future stock prices.
- Weather forecasting :
Temperature, humidity, wind speed, rainfall, sunlight are time-series data measured on a small-time span of minutes. By studying past inputs, the model can forecast the climatic details. Weather forecasting is crucial in cases of cyclones, earthquakes, floods. People’s migration and other climate management activities are highly dependent on the results of the model.
Different Forecasting applications in Company
In the figure below, we can see that Time Series analysis can help at every level the organization works. Sales and Profit forecasting before working on product development set a target. Employee churning, budget forecasting helps for cost analysis.
In a company, time-series forecasting can play a big role and can be used at multiple levels. Let’s try to understand this with the above diagram.
A company XYZ wants to build a new product. It already has a few products of its own. Profit forecasting tells about the profit company can expect in the coming quarters for their respective products. If the company has its own shares, stock market prediction plays a very important role in the growth of the company. Things like launching a new product create hype and it leads to people investing in company shares. Hence, new product launches and stocks go hand in hand. With new products in picture, we need a dynamic team, here is where employees and their churning plays a big role. Any successful company wants its key employees to stay and forecasting is important in employee retention. For the team, the budget will get sanctioned. This budget sanctioning is dependent on the profits, customer satisfaction, and the funding received by the company. In budgets as well forecasting can be used. Once the product is developed, sales forecasting comes into play. If the product can be sold online, making the servers ready for the traffic, the stock presence of the product is very important and forecasting is highly used here. Thus, in different functioning of the company, forecasting can be used.