Abstract
There are many implementations of machine learning algorithms and other artificial intelligence used in wide variety of tools and technologies in our world. Events in our daily life are subject to such machine learning to hopefully make tasks simpler for humans. Humans get to think less about mundane tasks, such as scheduling daily activities and machines get to “think” more.
This paper begins an exploration of an implementation for a simple and rather routine schedule. Titled Prof’s Personal Assistant, the programming portions first learn rules for scheduling a day, a time and duration for events that occur in this professor’s schedule. Obvious future features could be continually added.
Decision trees, in particular, ID3 and Quinlan’s C4.5 were used to create new rules for day, time and duration of events in a schedule. In the experiments decision tree were generated and rules were created based on one month of events, then performance statistics were gathered for one week’s worth of scheduling. Rules were generated again, based on the month and the new week’s information. A large component of personalizing a schedule includes gathering the information of the usage. This procedure of adding a week and then regenerating rules was applied to six weeks of scheduling, and therefore data gathering and statistical data gathering. Statistical data was represented for each rule as simple fractions on successful predictions to total number of predictions. This was used to compare the methods of rule generation.
This forgetful way of learning bases the learning only on the day and forgets all rules, even those that added to the improved performance of the value guessing.