I'm new to this community and I don't know if this is the right place to ask this question, sorry if I'm wrong. It's kind of a subjective question but I have no knowledge about this forecast type, the example helps me to explain the problem.
I would like to know what is the best approach to get forecasts with machine learning algorithms for my scenario. I have a dataset gave to me to play around by my company that concerns monthly effort registered by employees working on several projects, and my goal is to forecast the employees effort for the next period (like for the next month or next n months) based on past data.
I cannot share the dataset because of sensitive informations, but I can describe how it's composed.
Each record has several features about:
total effort registered for a specific project(made up of all employees effort)
specific employee effort for a project(which is a percentage of the project total effort)
Total Employee effort(sum of all of its effort in a specified month)
Number of projects an employee works for each month
Type of project(like marketing - production line - engineering)
Project milestone status(one phase between n different phases)
An example of record would be like:
"Hardest project ever"
20(sum for all employees of "Hardest project ever")
2(George effort for project "Hardest project ever")
5(Sum of George effort for project "Hardest project ever" and "the easiest project")
2(number of projects George works on for June 2019)
engineering("Hardest project ever" project type)
2(not finished, project milestone status)
Other records are stored in the dataset for each employee-time-project combinations, like for "the easiest project" and other employees (other than George) for June 2019.The dataset increases its size each month with new data.
I'm interested in forecasting for each employee what will be, based on past observations, his Total employee effort (6th in the list) for future months.
The problem is that I can't find what's the best approach, for this kind of forecast, that I could apply. I cannot think about Time Series forecasting (for each employee) because there are many employees and it would be really expensive to analyze time series one by one and apply ARIMA or similar models, too much time would be needed. I'm confident that something easier and faster exists.
Even if it's possible to use a simple Machine Learning Algorithm (NN, Ensemble models, KNN, SVM, ...), would be correct to use them on the whole original dataset in "regression mode" to get predictions? I have thought about splitting the dataset in train-test after ordering it by time keeping 1 year for test against 3 years of training, but I'm not sure that is the right way to get the job done.
All the forecast examples I can find on the internet are about single feature time series, not with more than one feature (I think that many features of the dataset could help to determine the target). Curerntly I'm using Python. Could someone give any useful advice?
PS: Is it wrong to call my dataset a "time-based" dataset? something like a Time Series is.