Machine learning is already alive! It's alive on Cross Validated and on Computer Science (plus Stack Overflow if you're programming).
Cross Validated has welcome machine learning questions from its inception. I don't see any evidence that CV is not a good home for machine learning questions. To my uninformed eyes, CV does have machine learning expertise. Sure, it isn't called “machine learning”, but this is not a problem. Experts on Java participate on Stack Overflow even though the site has a majority of non-Java content. Experts on differential topology participate on Mathematics even though the site has a majority of non-differential-topology content. Similarly, experts on machine learning can participate on Cross Validated even though the site isn't solely about ML. It's easy to focus on one tag if you don't care about non-ML topics.
This is the second time a machine learning proposal fails due to a lack of traffic. The first time over, the content was merged into Cross Validated. This time over, there were even fewer questions in the private beta. MLare is just too narrow a topic to accommodate a separate site, especially when there's already a thriving site. During the second ML private beta, about 46 ML questions were posted on Cross Validated, plus a few more on CS.
You can propose a new machine learning site on Area 51. It won't be automatically restarted: you have to work for it. But there's no point in investing any energy into that if you aren't going to do something different. It failed twice, so what makes you think the third time is going to succeed? And what's the point anyway — how would the site make the Internet a better place? You need to answer these questions first. If you have an answer to both, then propose away.