On urban roads, 25% of all car accidents involve pedestrians who enter the roadway. Drivers' uncertainty about safety and their inattentiveness account for 70% of these accidents. The purpose of this study is to analyze drivers' attention to pedestrians so as to give an alert when the driver's attention is insufficient. This method consists of estimating drivers' attention to pedestrians using gaze information, which is 90% of the cognitive information source when driving; operation information on the driver; and momentary change. A driver's gaze information was estimated based on information sensed by a gaze-measuring instrument. Braking and accelerating were used as driver operation information. The relationship between gazing target, operation information, momentary change amount, and contact with and avoidance of a pedestrian who jumped out was analyzed using a support vector machine (SVM). Evaluation was carried out by preparing a course simulating a city road on a driving simulator, asking a subject to drive the course, and using that data. Using this method, for a pedestrian who jumped into the roadway, we were able to predict whether the driver would collide or avoid with a probability of 70% when Time to Collision (TTC) was 4 seconds. We were able to confirm the possibility of alerting the driver to avoid accidents with pedestrians on urban roads.