Budi Arsana

If you want to code it yourself and based on your desired output. RNN (Recurrent neural network) would be the right choice. You can code it using Tensorflow and use LSTM as the layers.

If you prefer on cloud with tools ready to use and not much coding, Amazon DeepAR Forecasting looks sufficient.

I think you scan skip MongoDB for now and focussing on creating web component with Reactjs or Vue, I would also recommend to use TypeScript for type hinting support.

For styling, learn CSS first then upgrade to SASS/SCSS or LESS (pick one as mostly same concept) to make CSS more maintainable.

Also to improve your skill on both sectors, install linters if available. For TypeScipt, there are TSLint and for styling, i think there are Stylint. Linter will help you adapt to make a clean code and understand how other peoples usually styled their code.

The best way to make it more maintainable is by make your PHP Backend provide API like REST API or GraphQL that will be consumed by your React application. And since you will code React, i would recommend to use TypeScript as the compiler for their typehint support and less sub component install/setup for the compiler.

I prefer to use GitHub because their website speed is faster and the availability more reliable than the competitors, this is our top priority as GitHub is our core functionality we need to be able to operate in development. And since their new pricing make more sense and work for us as they are charging based on how many developers than repositories, this help us to keep our repositories smaller by dividing each codebase into specifics repositories.