1. This article describes how you can make your first Time Series model using Obviously AI. Specifically, we provide univariate time series forecasting. Time series problems are those that involve a time component. Time does play a role in standard AutoML datasets as well but a time series dataset is different. Time series adds an explicit order dependence between observations: a time dimension. A time series is a sequence of observations taken sequentially in time. We specifically focus on univariate time series forecasting, i.e., where we have a single numeric column that is time dependent

  2. However, please note that you can upload your dataset as it is, i.e., your dataset does not need to consist of only the date column and the prediction column. In case your dataset has multiple columns, you have the option to choose your prediction column after uploading it on the platform. This feature also gives you the ability to use multiple numeric columns as your prediction column one by one, if you are interested in performing time series analysis on multiple time dependent numeric columns present in your data

  3. First, within the platform we select the “+” button to upload a dataset

    Next, we choose the datasource. In this case we choose to upload a CSV file

    After clicking on the upload button, we get the screen with the dataset requirements on the left, and the option to upload/drag and drop your data onto the platform. Make sure that you toggle to Time Series to check the correct requirements and then fill in the required fields as shown below. Currently, we support minimum 12 rows and maximum 1 million rows of data. The date column needs to be in the correct format, however, the platform automatically converts standard date formats to our acceptable yyyy-mm-dd format

    After entering the details, click on the “Upload” button. Once the dataset is uploaded you will first see a review of your data, where the platform details the total number of rows and columns in the data, and the total number of unfit rows and columns. For each column it shows whether the column is fit for use, %of missing values in a column and its data type. On the right section you will also be able to see the graphical distribution of the data

    Now you can click on continue and you will be directed to the next section

    where you get the option of choosing the particular Date column (in case you have multiple date columns, else the platform automatically chooses the only date column). It is advisable to check for the date format and also sort the dates in ascending order before uploading the dataset. However the platform does automatically sort the dates as well when required

    Next is choosing the prediction column (in case you have multiple numeric columns in your dataset which are automatically filtered by the platform and can be seen through the dropdown associated with the prediction column, this is important because remember time series works with numeric columns only). In case you have only one numeric column, that will automatically be chosen as the prediction column by the platform

    On the right, you will now see the graph with Date on the x-axis and the prediction column on the y-axis. Thus you will have a fair understanding of the pattern of your data by looking into this graph. Also, you can hover over the graph and check the x,y values for each point on the graph

    Additionally, you will now have to choose the Data Level, Aggregation function and Seasonality of the data. By default the values are Month, Sum and 12 respectively. But make sure to use the drop down to choose the correct values for each, as this depends on your data. For example, Data Level can be Day/Week/Month/Quarter depending on what level is your dataset

    Aggregation functions can be either sum/average, you can choose whether you want to aggregate your data using sum/average, during aggregation gaps in the data are removed and then aggregated

    Seasonality refers to the pattern of your data over a year, for example any repetitions or predictable behavior that repeats every year. The recommended values for seasonality are as mentioned below.

    Day: 7/30/365

    Week: 52

    Month: 12

    Quarter: 4

    The car sales dataset that we have used here, has the data originally for every month, so we choose the data level, aggregation function and seasonality as Month, Sum and 12 respectively. That’s it, you next click on the Start Predicting button to start building your model.

    As soon as your prediction report is generated, you know that your first Time series model is built. You can edit the name of the report just by double clicking on the existing name.

Did this answer your question?