Predictive models that output discrete classes or categories are classification models. Classification is widely used in the real world for use cases such as sentiment analysis of text and identifying objects in images. In this course, you will review how classification models can be used to categorize or classify input records. You will learn how metrics such as accuracy, precision, and recall can be used to evaluate classification models and the conditions under which you would choose to use precision and recall over accuracy for model evaluation. Next, you will use the BigQuery command-line tool bq to create a BigQuery dataset and table and load data into that table. You will see how you can run queries and explore your data, all using the command line. You will use Looker Studio for data visualization and DataPrep to clean and prepare your classification data. Finally, you will train a binary classification model and a multi-class classification model. You will improve the model's performance by balancing the records in the different categories and by using hyperparameter tuning to find the best model for your data.

BigQuery is a flagship product on the Google Cloud Platform which allows you to build and train machine learning (ML) models using simple SQL queries. BigQuery has support for a range of supervised and unsupervised machine learning models that can be trained on data stored in BigQuery. In this course, you will be introduced to BigQuery on the Google Cloud Platform and set up a GCP trial account that allows you to work with BigQuery to train ML models. You will then review some machine learning basics and dig a little deeper into regression models. Next, you will create datasets and tables in BigQuery and upload your data to the cloud. You will visualize and explore your data using Looker Studio and prepare and clean your data using DataPrep. Finally, you will train regression models using linear regression, gradient-boosted trees, and the random forest model and evaluate and compare the performance of these models on your test data.

Unsupervised techniques such as clustering and recommendation systems can discover patterns in unlabeled data. These models extract structure in the x-variables or features present in the data. In this course, you will work with two unsupervised learning methods, clustering and recommendation systems. You will explore how clustering algorithms use only the x-variables or features in your data to group data into logical clusters. Then you will discover the basic concepts behind recommendation systems, which take in past user interactions with products and use that to recommend new products to users. Next, you will train a clustering model using k-means clustering on your data and evaluate how the clusters differ. You will use hyperparameter tuning to find the best number of clusters on your dataset. Finally, you will train a recommendations engine using collaborative filtering and use that to make movie recommendations to users based on their past preferences and the preferences of other users.

Time series forecasting uses data collected over periodic intervals to understand and analyze how the variable changes over time. Time series analysis is used for forecasting problems, such as demand forecasting and revenue forecasting. The auto-regressive integrated moving average (ARIMA) model is widely used for time series forecasting. In this course, you will see how time series analysis works and how models such as the ARIMA model can help you forecast future values of time-varying data using historical values. You will also learn the differences between stationary and non-stationary time series data. Next, you will load and explore your time series data for store revenue prediction into BigQuery and visualize and explore this data using Looker Studio. Finally, you will use an ARIMA model to make revenue forecasts. You will see how BigQuery ML trains multiple ARIMA models to find the best auto-regressive, differencing, and moving average parameters for your data. You will also perform multiple time-series analysis by forecasting store revenue by region.