Open-AudIT v2.0 (OAv2) does exactly this.
http://www.open-audit.org

Disclaimer – I am the developer. I am/was a SysAdmin who (still) runs into this exact issue. If you “audit” your systems each day, it will create a report that says in the next XX days, server ABC, partition 123 will reach 100% capacity, based on past useage stats and trends.

My primary point is that it’s high time move on from the 1990′s style monitoring that we’re still doing, and start applying some more maths. The statistical field of prediction is well over 40 years old. Many of the known algorithms aren’t even particularly computationally intensive. There’s really no excuse.

Another thing is possible interrelated metrics. An obvious example: When physical memory utilisation goes beyond 100% and pages start getting swapped out, disk I/O is likely to increase dramatically. Predicting the trend of one metric might benefit from a predicted trend of another metric.

The possibilities are vast. One could collect data points on a global scale and mine it for association rules and correlations. Or, one could make do with the data available locally and discover interdependencies dynamically. Even a few handcrafted rules would probably do relatively well.

Also, filesystem utilisation is a very simplistic use case. It’s almost universally increasing, and quite often at a steady, constant pace. A simple linear regression will probably fail against data that exhibits more erratic pattern with a long-term linear trend or if the data set has a non-linear trend.

[1] http://oss.oetiker.ch/rrdtool/gallery/index.en.html