To explore the fatigue characteristics of novice drivers during conditional automated driving (CAD), this study analyzed the impact of different types of non-driving-related tasks (NDRTs) (watching videos, resting, monitoring operations, reading texts) and durations (5, 10, 20, 30 min) on driver fatigue. Various scenarios with different NDRT types and durations were designed using a driving simulation platform. Thirty-five young novice drivers participated, and Karolinska Sleepiness Scale (KSS) and electroencephalogram (EEG) data were collected. A Long Short-Term Memory (LSTM) neural network model was used to impute the KSS data during rest tasks. LSTM neural networks estimated KSS data during rest tasks with about 94% accuracy. Results indicated that monitoring operations caused the highest fatigue, followed by reading texts, watching videos, and resting. Fatigue increased with task duration, with the most significant rise in KSS observed during the initial 5 min for resting and 15 min for other NDRTs. The takeover process notably reduced KSS by 1.7, though manual driving was minimally affected. No significant correlation existed between accidents during takeovers and KSS, NDRT types, or durations. The findings highlight the importance of customizing fatigue management strategies based on the type and duration of NDRTs to enhance safety in CAD environments.